1
|
Hu C, Hasenstaub AR, Schreiner CE. Basic Properties of Coordinated Neuronal Ensembles in the Auditory Thalamus. J Neurosci 2024; 44:e1729232024. [PMID: 38561224 PMCID: PMC11079962 DOI: 10.1523/jneurosci.1729-23.2024] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/13/2023] [Revised: 03/04/2024] [Accepted: 03/11/2024] [Indexed: 04/04/2024] Open
Abstract
Coordinated neuronal activity has been identified to play an important role in information processing and transmission in the brain. However, current research predominantly focuses on understanding the properties and functions of neuronal coordination in hippocampal and cortical areas, leaving subcortical regions relatively unexplored. In this study, we use single-unit recordings in female Sprague Dawley rats to investigate the properties and functions of groups of neurons exhibiting coordinated activity in the auditory thalamus-the medial geniculate body (MGB). We reliably identify coordinated neuronal ensembles (cNEs), which are groups of neurons that fire synchronously, in the MGB. cNEs are shown not to be the result of false-positive detections or by-products of slow-state oscillations in anesthetized animals. We demonstrate that cNEs in the MGB have enhanced information-encoding properties over individual neurons. Their neuronal composition is stable between spontaneous and evoked activity, suggesting limited stimulus-induced ensemble dynamics. These MGB cNE properties are similar to what is observed in cNEs in the primary auditory cortex (A1), suggesting that ensembles serve as a ubiquitous mechanism for organizing local networks and play a fundamental role in sensory processing within the brain.
Collapse
Affiliation(s)
- Congcong Hu
- John & Edward Coleman Memorial Laboratory, University of California-San Francisco, San Francisco, California 94158
- Neuroscience Graduate Program, University of California-San Francisco, San Francisco, California 94158
- Department of Otolaryngology-Head and Neck Surgery, University of California-San Francisco, San Francisco, California 94158
| | - Andrea R Hasenstaub
- John & Edward Coleman Memorial Laboratory, University of California-San Francisco, San Francisco, California 94158
- Neuroscience Graduate Program, University of California-San Francisco, San Francisco, California 94158
- Department of Otolaryngology-Head and Neck Surgery, University of California-San Francisco, San Francisco, California 94158
| | - Christoph E Schreiner
- John & Edward Coleman Memorial Laboratory, University of California-San Francisco, San Francisco, California 94158
- Neuroscience Graduate Program, University of California-San Francisco, San Francisco, California 94158
- Department of Otolaryngology-Head and Neck Surgery, University of California-San Francisco, San Francisco, California 94158
| |
Collapse
|
2
|
Lindeberg T. A time-causal and time-recursive scale-covariant scale-space representation of temporal signals and past time. BIOLOGICAL CYBERNETICS 2023; 117:21-59. [PMID: 36689001 PMCID: PMC10160219 DOI: 10.1007/s00422-022-00953-6] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Received: 02/18/2022] [Accepted: 11/21/2022] [Indexed: 05/05/2023]
Abstract
This article presents an overview of a theory for performing temporal smoothing on temporal signals in such a way that: (i) temporally smoothed signals at coarser temporal scales are guaranteed to constitute simplifications of corresponding temporally smoothed signals at any finer temporal scale (including the original signal) and (ii) the temporal smoothing process is both time-causal and time-recursive, in the sense that it does not require access to future information and can be performed with no other temporal memory buffer of the past than the resulting smoothed temporal scale-space representations themselves. For specific subsets of parameter settings for the classes of linear and shift-invariant temporal smoothing operators that obey this property, it is shown how temporal scale covariance can be additionally obtained, guaranteeing that if the temporal input signal is rescaled by a uniform temporal scaling factor, then also the resulting temporal scale-space representations of the rescaled temporal signal will constitute mere rescalings of the temporal scale-space representations of the original input signal, complemented by a shift along the temporal scale dimension. The resulting time-causal limit kernel that obeys this property constitutes a canonical temporal kernel for processing temporal signals in real-time scenarios when the regular Gaussian kernel cannot be used, because of its non-causal access to information from the future, and we cannot additionally require the temporal smoothing process to comprise a complementary memory of the past beyond the information contained in the temporal smoothing process itself, which in this way also serves as a multi-scale temporal memory of the past. We describe how the time-causal limit kernel relates to previously used temporal models, such as Koenderink's scale-time kernels and the ex-Gaussian kernel. We do also give an overview of how the time-causal limit kernel can be used for modelling the temporal processing in models for spatio-temporal and spectro-temporal receptive fields, and how it more generally has a high potential for modelling neural temporal response functions in a purely time-causal and time-recursive way, that can also handle phenomena at multiple temporal scales in a theoretically well-founded manner. We detail how this theory can be efficiently implemented for discrete data, in terms of a set of recursive filters coupled in cascade. Hence, the theory is generally applicable for both: (i) modelling continuous temporal phenomena over multiple temporal scales and (ii) digital processing of measured temporal signals in real time. We conclude by stating implications of the theory for modelling temporal phenomena in biological, perceptual, neural and memory processes by mathematical models, as well as implications regarding the philosophy of time and perceptual agents. Specifically, we propose that for A-type theories of time, as well as for perceptual agents, the notion of a non-infinitesimal inner temporal scale of the temporal receptive fields has to be included in representations of the present, where the inherent nonzero temporal delay of such time-causal receptive fields implies a need for incorporating predictions from the actual time-delayed present in the layers of a perceptual hierarchy, to make it possible for a representation of the perceptual present to constitute a representation of the environment with timing properties closer to the actual present.
Collapse
Affiliation(s)
- Tony Lindeberg
- Computational Brain Science Lab, Division of Computational Science and Technology, KTH Royal Institute of Technology, 100 44, Stockholm, Sweden.
| |
Collapse
|
3
|
He F, Stevenson IH, Escabí MA. Two stages of bandwidth scaling drives efficient neural coding of natural sounds. PLoS Comput Biol 2023; 19:e1010862. [PMID: 36787338 PMCID: PMC9970106 DOI: 10.1371/journal.pcbi.1010862] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/28/2022] [Revised: 02/27/2023] [Accepted: 01/09/2023] [Indexed: 02/15/2023] Open
Abstract
Theories of efficient coding propose that the auditory system is optimized for the statistical structure of natural sounds, yet the transformations underlying optimal acoustic representations are not well understood. Using a database of natural sounds including human speech and a physiologically-inspired auditory model, we explore the consequences of peripheral (cochlear) and mid-level (auditory midbrain) filter tuning transformations on the representation of natural sound spectra and modulation statistics. Whereas Fourier-based sound decompositions have constant time-frequency resolution at all frequencies, cochlear and auditory midbrain filters bandwidths increase proportional to the filter center frequency. This form of bandwidth scaling produces a systematic decrease in spectral resolution and increase in temporal resolution with increasing frequency. Here we demonstrate that cochlear bandwidth scaling produces a frequency-dependent gain that counteracts the tendency of natural sound power to decrease with frequency, resulting in a whitened output representation. Similarly, bandwidth scaling in mid-level auditory filters further enhances the representation of natural sounds by producing a whitened modulation power spectrum (MPS) with higher modulation entropy than both the cochlear outputs and the conventional Fourier MPS. These findings suggest that the tuning characteristics of the peripheral and mid-level auditory system together produce a whitened output representation in three dimensions (frequency, temporal and spectral modulation) that reduces redundancies and allows for a more efficient use of neural resources. This hierarchical multi-stage tuning strategy is thus likely optimized to extract available information and may underlies perceptual sensitivity to natural sounds.
Collapse
Affiliation(s)
- Fengrong He
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
| | - Ian H. Stevenson
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- Psychological Sciences, University of Connecticut, Storrs, Connecticut, United States of America
- The Connecticut Institute for Brain and Cognitive Sciences, University of Connecticut, Storrs, Connecticut, United States of America
| | - Monty A. Escabí
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- Psychological Sciences, University of Connecticut, Storrs, Connecticut, United States of America
- The Connecticut Institute for Brain and Cognitive Sciences, University of Connecticut, Storrs, Connecticut, United States of America
- Electrical and Computer Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- * E-mail:
| |
Collapse
|
4
|
Gilday OD, Praegel B, Maor I, Cohen T, Nelken I, Mizrahi A. Surround suppression in mouse auditory cortex underlies auditory edge detection. PLoS Comput Biol 2023; 19:e1010861. [PMID: 36656876 PMCID: PMC9888713 DOI: 10.1371/journal.pcbi.1010861] [Citation(s) in RCA: 1] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/11/2022] [Revised: 01/31/2023] [Accepted: 01/09/2023] [Indexed: 01/20/2023] Open
Abstract
Surround suppression (SS) is a fundamental property of sensory processing throughout the brain. In the auditory system, the early processing stream encodes sounds using a one dimensional physical space-frequency. Previous studies in the auditory system have shown SS to manifest as bandwidth tuning around the preferred frequency. We asked whether bandwidth tuning can be found around frequencies away from the preferred frequency. We exploited the simplicity of spectral representation of sounds to study SS by manipulating both sound frequency and bandwidth. We recorded single unit spiking activity from the auditory cortex (ACx) of awake mice in response to an array of broadband stimuli with varying central frequencies and bandwidths. Our recordings revealed that a significant portion of neuronal response profiles had a preferred bandwidth that varied in a regular way with the sound's central frequency. To gain insight into the possible mechanism underlying these responses, we modelled neuronal activity using a variation of the "Mexican hat" function often used to model SS. The model accounted for response properties of single neurons with high accuracy. Our data and model show that these responses in ACx obey simple rules resulting from the presence of lateral inhibitory sidebands, mostly above the excitatory band of the neuron, that result in sensitivity to the location of top frequency edges, invariant to other spectral attributes. Our work offers a simple explanation for auditory edge detection and possibly other computations of spectral content in sounds.
Collapse
Affiliation(s)
- Omri David Gilday
- The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel
| | - Benedikt Praegel
- The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel
- Department of Neurobiology, The Hebrew University of Jerusalem, Jerusalem, Israel
| | - Ido Maor
- The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel
- Department of Neurobiology, The Hebrew University of Jerusalem, Jerusalem, Israel
| | - Tav Cohen
- Department of Neurobiology, The Hebrew University of Jerusalem, Jerusalem, Israel
| | - Israel Nelken
- The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel
- Department of Neurobiology, The Hebrew University of Jerusalem, Jerusalem, Israel
| | - Adi Mizrahi
- The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel
- Department of Neurobiology, The Hebrew University of Jerusalem, Jerusalem, Israel
- * E-mail:
| |
Collapse
|
5
|
Karbasi M, Kolossa D. ASR-based speech intelligibility prediction: A review. Hear Res 2022; 426:108606. [PMID: 36154977 DOI: 10.1016/j.heares.2022.108606] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 12/01/2021] [Revised: 08/15/2022] [Accepted: 09/12/2022] [Indexed: 11/04/2022]
Abstract
Various types of methods and approaches are available to predict the intelligibility of speech signals, but many of these still suffer from two major problems: first, their required prior knowledge, which itself could limit the applicability and lower the objectivity of the method, and second, a low generalization capacity, e.g. across noise types, degradation conditions, and speech material. Automatic speech recognition (ASR) has been suggested as a machine-learning-based component of speech intelligibility prediction (SIP), aiming to ameliorate the shortcomings of other SIP methods. Since their first introduction, ASR-based SIP approaches have been developing at an increasingly rapid pace, were deployed in a range of contexts, and have shown promising performance in many scenarios. Our article provides an overview of this body of research. The main differences between competing methods are highlighted and their benefits are explained next to their limitations. We conclude with an outlook on future work and new related directions.
Collapse
Affiliation(s)
- Mahdie Karbasi
- Cognitive Signal Processing Group, Faculty of Electrical Engineering and Information Technology, Ruhr University Bochum, 44801, NRW, Germany.
| | - Dorothea Kolossa
- Cognitive Signal Processing Group, Faculty of Electrical Engineering and Information Technology, Ruhr University Bochum, 44801, NRW, Germany
| |
Collapse
|
6
|
Dehghani A, Seyyedsalehi SA. Time-Frequency Localization Using Deep Convolutional Maxout Neural Network in Persian Speech Recognition. Neural Process Lett 2022. [DOI: 10.1007/s11063-022-11006-1] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/15/2022]
|
7
|
Task Engagement Improves Neural Discriminability in the Auditory Midbrain of the Marmoset Monkey. J Neurosci 2020; 41:284-297. [PMID: 33208469 DOI: 10.1523/jneurosci.1112-20.2020] [Citation(s) in RCA: 11] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/07/2020] [Revised: 10/24/2020] [Accepted: 10/27/2020] [Indexed: 11/21/2022] Open
Abstract
While task-dependent changes have been demonstrated in auditory cortex for a number of behavioral paradigms and mammalian species, less is known about how behavioral state can influence neural coding in the midbrain areas that provide auditory information to cortex. We measured single-unit activity in the inferior colliculus (IC) of common marmosets of both sexes while they performed a tone-in-noise detection task and during passive presentation of identical task stimuli. In contrast to our previous study in the ferret IC, task engagement had little effect on sound-evoked activity in central (lemniscal) IC of the marmoset. However, activity was significantly modulated in noncentral fields, where responses were selectively enhanced for the target tone relative to the distractor noise. This led to an increase in neural discriminability between target and distractors. The results confirm that task engagement can modulate sound coding in the auditory midbrain, and support a hypothesis that subcortical pathways can mediate highly trained auditory behaviors.SIGNIFICANCE STATEMENT While the cerebral cortex is widely viewed as playing an essential role in the learning and performance of complex auditory behaviors, relatively little attention has been paid to the role of brainstem and midbrain areas that process sound information before it reaches cortex. This study demonstrates that the auditory midbrain is also modulated during behavior. These modulations amplify task-relevant sensory information, a process that is traditionally attributed to cortex.
Collapse
|
8
|
Spiking network optimized for word recognition in noise predicts auditory system hierarchy. PLoS Comput Biol 2020; 16:e1007558. [PMID: 32559204 PMCID: PMC7329140 DOI: 10.1371/journal.pcbi.1007558] [Citation(s) in RCA: 8] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/29/2019] [Revised: 07/01/2020] [Accepted: 11/22/2019] [Indexed: 11/21/2022] Open
Abstract
The auditory neural code is resilient to acoustic variability and capable of recognizing sounds amongst competing sound sources, yet, the transformations enabling noise robust abilities are largely unknown. We report that a hierarchical spiking neural network (HSNN) optimized to maximize word recognition accuracy in noise and multiple talkers predicts organizational hierarchy of the ascending auditory pathway. Comparisons with data from auditory nerve, midbrain, thalamus and cortex reveals that the optimal HSNN predicts several transformations of the ascending auditory pathway including a sequential loss of temporal resolution and synchronization ability, increasing sparseness, and selectivity. The optimal organizational scheme enhances performance by selectively filtering out noise and fast temporal cues such as voicing periodicity, that are not directly relevant to the word recognition task. An identical network arranged to enable high information transfer fails to predict auditory pathway organization and has substantially poorer performance. Furthermore, conventional single-layer linear and nonlinear receptive field networks that capture the overall feature extraction of the HSNN fail to achieve similar performance. The findings suggest that the auditory pathway hierarchy and its sequential nonlinear feature extraction computations enhance relevant cues while removing non-informative sources of noise, thus enhancing the representation of sounds in noise impoverished conditions. The brain’s ability to recognize sounds in the presence of competing sounds or background noise is essential for everyday hearing tasks. How the brain accomplishes noise resiliency, however, is poorly understood. Using neural recordings from the ascending auditory pathway and an auditory spiking network model trained for sound recognition in noise we explore the computational strategies that enable noise robustness. Our results suggest that the hierarchical feature organization of the ascending auditory pathway and the resulting computations are critical for sound recognition in the presence of noise.
Collapse
|
9
|
Robust Rate-Place Coding of Resolved Components in Harmonic and Inharmonic Complex Tones in Auditory Midbrain. J Neurosci 2020; 40:2080-2093. [PMID: 31996454 DOI: 10.1523/jneurosci.2337-19.2020] [Citation(s) in RCA: 8] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/25/2019] [Revised: 01/12/2020] [Accepted: 01/16/2020] [Indexed: 11/21/2022] Open
Abstract
Harmonic complex tones (HCTs) commonly occurring in speech and music evoke a strong pitch at their fundamental frequency (F0), especially when they contain harmonics individually resolved by the cochlea. When all frequency components of an HCT are shifted by the same amount, the pitch of the resulting inharmonic tone (IHCT) can also shift, although the envelope repetition rate is unchanged. A rate-place code, whereby resolved harmonics are represented by local maxima in firing rates along the tonotopic axis, has been characterized in the auditory nerve and primary auditory cortex, but little is known about intermediate processing stages. We recorded single-neuron responses to HCT and IHCT with varying F0 and sound level in the inferior colliculus (IC) of unanesthetized rabbits of both sexes. Many neurons showed peaks in firing rate when a low-numbered harmonic aligned with the neuron's characteristic frequency, demonstrating "rate-place" coding. The IC rate-place code was most prevalent for F0 > 800 Hz, was only moderately dependent on sound level over a 40 dB range, and was not sensitive to stimulus harmonicity. A spectral receptive-field model incorporating broadband inhibition better predicted the neural responses than a purely excitatory model, suggesting an enhancement of the rate-place representation by inhibition. Some IC neurons showed facilitation in response to HCT relative to pure tones, similar to cortical "harmonic template neurons" (Feng and Wang, 2017), but to a lesser degree. Our findings shed light on the transformation of rate-place coding of resolved harmonics along the auditory pathway.SIGNIFICANCE STATEMENT Harmonic complex tones are ubiquitous in speech and music and produce strong pitch percepts when they contain frequency components that are individually resolved by the cochlea. Here, we characterize a "rate-place" code for resolved harmonics in the auditory midbrain that is more robust across sound levels than the peripheral rate-place code and insensitive to the harmonic relationships among frequency components. We use a computational model to show that inhibition may play an important role in shaping the rate-place code. Our study fills a major gap in understanding the transformations in neural representations of resolved harmonics along the auditory pathway.
Collapse
|
10
|
Dodds EM, DeWeese MR. On the Sparse Structure of Natural Sounds and Natural Images: Similarities, Differences, and Implications for Neural Coding. Front Comput Neurosci 2019; 13:39. [PMID: 31293408 PMCID: PMC6606779 DOI: 10.3389/fncom.2019.00039] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/28/2018] [Accepted: 06/05/2019] [Indexed: 11/25/2022] Open
Abstract
Sparse coding models of natural images and sounds have been able to predict several response properties of neurons in the visual and auditory systems. While the success of these models suggests that the structure they capture is universal across domains to some degree, it is not yet clear which aspects of this structure are universal and which vary across sensory modalities. To address this, we fit complete and highly overcomplete sparse coding models to natural images and spectrograms of speech and report on differences in the statistics learned by these models. We find several types of sparse features in natural images, which all appear in similar, approximately Laplace distributions, whereas the many types of sparse features in speech exhibit a broad range of sparse distributions, many of which are highly asymmetric. Moreover, individual sparse coding units tend to exhibit higher lifetime sparseness for overcomplete models trained on images compared to those trained on speech. Conversely, population sparseness tends to be greater for these networks trained on speech compared with sparse coding models of natural images. To illustrate the relevance of these findings to neural coding, we studied how they impact a biologically plausible sparse coding network's representations in each sensory modality. In particular, a sparse coding network with synaptically local plasticity rules learns different sparse features from speech data than are found by more conventional sparse coding algorithms, but the learned features are qualitatively the same for these models when trained on natural images.
Collapse
Affiliation(s)
- Eric McVoy Dodds
- Redwood Center for Theoretical Neuroscience, University of California, Berkeley, Berkeley, CA, United States
- Department of Physics, University of California, Berkeley, Berkeley, CA, United States
| | - Michael Robert DeWeese
- Redwood Center for Theoretical Neuroscience, University of California, Berkeley, Berkeley, CA, United States
- Department of Physics, University of California, Berkeley, Berkeley, CA, United States
- Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, United States
| |
Collapse
|
11
|
Daube C, Ince RAA, Gross J. Simple Acoustic Features Can Explain Phoneme-Based Predictions of Cortical Responses to Speech. Curr Biol 2019; 29:1924-1937.e9. [PMID: 31130454 PMCID: PMC6584359 DOI: 10.1016/j.cub.2019.04.067] [Citation(s) in RCA: 69] [Impact Index Per Article: 13.8] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/22/2018] [Revised: 03/25/2019] [Accepted: 04/25/2019] [Indexed: 01/06/2023]
Abstract
When we listen to speech, we have to make sense of a waveform of sound pressure. Hierarchical models of speech perception assume that, to extract semantic meaning, the signal is transformed into unknown, intermediate neuronal representations. Traditionally, studies of such intermediate representations are guided by linguistically defined concepts, such as phonemes. Here, we argue that in order to arrive at an unbiased understanding of the neuronal responses to speech, we should focus instead on representations obtained directly from the stimulus. We illustrate our view with a data-driven, information theoretic analysis of a dataset of 24 young, healthy humans who listened to a 1 h narrative while their magnetoencephalogram (MEG) was recorded. We find that two recent results, the improved performance of an encoding model in which annotated linguistic and acoustic features were combined and the decoding of phoneme subgroups from phoneme-locked responses, can be explained by an encoding model that is based entirely on acoustic features. These acoustic features capitalize on acoustic edges and outperform Gabor-filtered spectrograms, which can explicitly describe the spectrotemporal characteristics of individual phonemes. By replicating our results in publicly available electroencephalography (EEG) data, we conclude that models of brain responses based on linguistic features can serve as excellent benchmarks. However, we believe that in order to further our understanding of human cortical responses to speech, we should also explore low-level and parsimonious explanations for apparent high-level phenomena.
Collapse
Affiliation(s)
- Christoph Daube
- Institute of Neuroscience and Psychology, University of Glasgow, 62 Hillhead Street, Glasgow G12 8QB, UK.
| | - Robin A A Ince
- Institute of Neuroscience and Psychology, University of Glasgow, 62 Hillhead Street, Glasgow G12 8QB, UK
| | - Joachim Gross
- Institute of Neuroscience and Psychology, University of Glasgow, 62 Hillhead Street, Glasgow G12 8QB, UK; Institute for Biomagnetism and Biosignalanalysis, University of Münster, Malmedyweg 15, 48149 Münster, Germany
| |
Collapse
|
12
|
Zhang Q, Hu X, Hong B, Zhang B. A hierarchical sparse coding model predicts acoustic feature encoding in both auditory midbrain and cortex. PLoS Comput Biol 2019; 15:e1006766. [PMID: 30742609 PMCID: PMC6386396 DOI: 10.1371/journal.pcbi.1006766] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.2] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/29/2018] [Revised: 02/22/2019] [Accepted: 12/21/2018] [Indexed: 12/03/2022] Open
Abstract
The auditory pathway consists of multiple stages, from the cochlear nucleus to the auditory cortex. Neurons acting at different stages have different functions and exhibit different response properties. It is unclear whether these stages share a common encoding mechanism. We trained an unsupervised deep learning model consisting of alternating sparse coding and max pooling layers on cochleogram-filtered human speech. Evaluation of the response properties revealed that computing units in lower layers exhibited spectro-temporal receptive fields (STRFs) similar to those of inferior colliculus neurons measured in physiological experiments, including properties such as sound onset and termination, checkerboard pattern, and spectral motion. Units in upper layers tended to be tuned to phonetic features such as plosivity and nasality, resembling the results of field recording in human auditory cortex. Variation of the sparseness level of the units in each higher layer revealed a positive correlation between the sparseness level and the strength of phonetic feature encoding. The activities of the units in the top layer, but not other layers, correlated with the dynamics of the first two formants (F1, F2) of all phonemes, indicating the encoding of phoneme dynamics in these units. These results suggest that the principles of sparse coding and max pooling may be universal in the human auditory pathway. When speech enters the ear, it is subjected to a series of processing stages prior to arriving at the auditory cortex. Neurons acting at different processing stages have different response properties. For example, at the auditory midbrain, a neuron may specifically detect the onsets of a frequency component in the speech, whereas in the auditory cortex, a neuron may specifically detect phonetic features. The encoding mechanisms underlying these neuronal functions remain unclear. To address this issue, we designed a hierarchical sparse coding model, inspired by the sparse activity of neurons in the sensory system, to learn features in speech signals. We found that the computing units in different layers exhibited hierarchical extraction of speech sound features, similar to those of neurons in the auditory midbrain and auditory cortex, although the computational principles in these layers were the same. The results suggest that sparse coding and max pooling represent universal computational principles throughout the auditory pathway.
Collapse
Affiliation(s)
- Qingtian Zhang
- Department of Computer Science and Technology, Tsinghua University, Beijing, China
| | - Xiaolin Hu
- Department of Computer Science and Technology, Tsinghua University, Beijing, China
- Center for Brain-Inspired Computing Research (CBICR), Tsinghua University, Beijing, China
- * E-mail:
| | - Bo Hong
- School of Medicine, Tsinghua University, Beijing, China
| | - Bo Zhang
- Department of Computer Science and Technology, Tsinghua University, Beijing, China
- Center for Brain-Inspired Computing Research (CBICR), Tsinghua University, Beijing, China
| |
Collapse
|
13
|
Spille C, Kollmeier B, Meyer BT. Comparing human and automatic speech recognition in simple and complex acoustic scenes. COMPUT SPEECH LANG 2018. [DOI: 10.1016/j.csl.2018.04.003] [Citation(s) in RCA: 15] [Impact Index Per Article: 2.5] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/29/2022]
|
14
|
Khatami F, Wöhr M, Read HL, Escabí MA. Origins of scale invariance in vocalization sequences and speech. PLoS Comput Biol 2018; 14:e1005996. [PMID: 29659561 PMCID: PMC5919684 DOI: 10.1371/journal.pcbi.1005996] [Citation(s) in RCA: 7] [Impact Index Per Article: 1.2] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/18/2017] [Revised: 04/26/2018] [Accepted: 01/23/2018] [Indexed: 11/18/2022] Open
Abstract
To communicate effectively animals need to detect temporal vocalization cues that vary over several orders of magnitude in their amplitude and frequency content. This large range of temporal cues is evident in the power-law scale-invariant relationship between the power of temporal fluctuations in sounds and the sound modulation frequency (f). Though various forms of scale invariance have been described for natural sounds, the origins and implications of scale invariant phenomenon remain unknown. Using animal vocalization sequences, including continuous human speech, and a stochastic model of temporal amplitude fluctuations we demonstrate that temporal acoustic edges are the primary acoustic cue accounting for the scale invariant phenomenon. The modulation spectrum of vocalization sequences and the model both exhibit a dual regime lowpass structure with a flat region at low modulation frequencies and scale invariant 1/f2 trend for high modulation frequencies. Moreover, we find a time-frequency tradeoff between the average vocalization duration of each vocalization sequence and the cutoff frequency beyond which scale invariant behavior is observed. These results indicate that temporal edges are universal features responsible for scale invariance in vocalized sounds. This is significant since temporal acoustic edges are salient perceptually and the auditory system could exploit such statistical regularities to minimize redundancies and generate compact neural representations of vocalized sounds. The efficient coding hypothesis posits that the brain encodes sensory signals efficiently in order to reduce metabolic cost and preserve behaviorally relevant environment information. In audition, recognition and coding depends on the brain’s ability to accurately and efficiently encode statistical regularities that are prevalent in natural sounds. Similarly, efficient audio coding and compression schemes attempt to preserve salient sound qualities while minimizing data bandwidth. A widely observed statistical regularity in nearly all natural sounds is the presence of scale invariance where the power of amplitude fluctuations is inversely related to the sound amplitude modulation frequency. In this study, we explore the physical sound cues responsible for the scale invariant phenomenon previously observed. We demonstrate that for animal vocalizations, including human speech, the scale invariant behavior is fully accounted by the presence of temporal acoustic edges that are largely created by opening and closing of the oral cavity and which mark the beginning and end of isolated vocalizations. The findings thus identify a single physical cue responsible for the universal scale invariant phenomenon that the brain can exploit to optimize coding and perception of vocalized sounds.
Collapse
Affiliation(s)
- Fatemeh Khatami
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
| | - Markus Wöhr
- Behavioral Neuroscience, Experimental and Biological Psychology, Faculty of Psychology, Philipps-University of Marburg, Marburg, Germany
- Center for Mind, Brain, and Behavior (CMBB), Philipps-University of Marburg, Marburg, Germany
| | - Heather L. Read
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- Department of Psychological Sciences, University of Connecticut, Storrs, Connecticut, United States of America
| | - Monty A. Escabí
- Biomedical Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- Department of Psychological Sciences, University of Connecticut, Storrs, Connecticut, United States of America
- Electrical and Computer Engineering, University of Connecticut, Storrs, Connecticut, United States of America
- * E-mail:
| |
Collapse
|
15
|
Holdgraf CR, Rieger JW, Micheli C, Martin S, Knight RT, Theunissen FE. Encoding and Decoding Models in Cognitive Electrophysiology. Front Syst Neurosci 2017; 11:61. [PMID: 29018336 PMCID: PMC5623038 DOI: 10.3389/fnsys.2017.00061] [Citation(s) in RCA: 65] [Impact Index Per Article: 9.3] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/07/2017] [Accepted: 08/07/2017] [Indexed: 11/13/2022] Open
Abstract
Cognitive neuroscience has seen rapid growth in the size and complexity of data recorded from the human brain as well as in the computational tools available to analyze this data. This data explosion has resulted in an increased use of multivariate, model-based methods for asking neuroscience questions, allowing scientists to investigate multiple hypotheses with a single dataset, to use complex, time-varying stimuli, and to study the human brain under more naturalistic conditions. These tools come in the form of "Encoding" models, in which stimulus features are used to model brain activity, and "Decoding" models, in which neural features are used to generated a stimulus output. Here we review the current state of encoding and decoding models in cognitive electrophysiology and provide a practical guide toward conducting experiments and analyses in this emerging field. Our examples focus on using linear models in the study of human language and audition. We show how to calculate auditory receptive fields from natural sounds as well as how to decode neural recordings to predict speech. The paper aims to be a useful tutorial to these approaches, and a practical introduction to using machine learning and applied statistics to build models of neural activity. The data analytic approaches we discuss may also be applied to other sensory modalities, motor systems, and cognitive systems, and we cover some examples in these areas. In addition, a collection of Jupyter notebooks is publicly available as a complement to the material covered in this paper, providing code examples and tutorials for predictive modeling in python. The aim is to provide a practical understanding of predictive modeling of human brain data and to propose best-practices in conducting these analyses.
Collapse
Affiliation(s)
- Christopher R. Holdgraf
- Department of Psychology, Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, United States
- Office of the Vice Chancellor for Research, Berkeley Institute for Data Science, University of California, Berkeley, Berkeley, CA, United States
| | - Jochem W. Rieger
- Department of Psychology, Carl-von-Ossietzky University, Oldenburg, Germany
| | - Cristiano Micheli
- Department of Psychology, Carl-von-Ossietzky University, Oldenburg, Germany
- Institut des Sciences Cognitives Marc Jeannerod, Lyon, France
| | - Stephanie Martin
- Department of Psychology, Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, United States
- Defitech Chair in Brain-Machine Interface, Center for Neuroprosthetics, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland
| | - Robert T. Knight
- Department of Psychology, Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, United States
| | - Frederic E. Theunissen
- Department of Psychology, Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA, United States
- Department of Psychology, University of California, Berkeley, Berkeley, CA, United States
| |
Collapse
|
16
|
Castro Martinez AM, Mallidi SH, Meyer BT. On the relevance of auditory-based Gabor features for deep learning in robust speech recognition. COMPUT SPEECH LANG 2017. [DOI: 10.1016/j.csl.2017.02.006] [Citation(s) in RCA: 7] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/20/2022]
|
17
|
Bach JH, Kollmeier B, Anemüller J. Matching Pursuit Analysis of Auditory Receptive Fields' Spectro-Temporal Properties. Front Syst Neurosci 2017; 11:4. [PMID: 28232791 PMCID: PMC5299023 DOI: 10.3389/fnsys.2017.00004] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/01/2016] [Accepted: 01/23/2017] [Indexed: 11/13/2022] Open
Abstract
Gabor filters have long been proposed as models for spectro-temporal receptive fields (STRFs), with their specific spectral and temporal rate of modulation qualitatively replicating characteristics of STRF filters estimated from responses to auditory stimuli in physiological data. The present study builds on the Gabor-STRF model by proposing a methodology to quantitatively decompose STRFs into a set of optimally matched Gabor filters through matching pursuit, and by quantitatively evaluating spectral and temporal characteristics of STRFs in terms of the derived optimal Gabor-parameters. To summarize a neuron's spectro-temporal characteristics, we introduce a measure for the “diagonality,” i.e., the extent to which an STRF exhibits spectro-temporal transients which cannot be factorized into a product of a spectral and a temporal modulation. With this methodology, it is shown that approximately half of 52 analyzed zebra finch STRFs can each be well approximated by a single Gabor or a linear combination of two Gabor filters. Moreover, the dominant Gabor functions tend to be oriented either in the spectral or in the temporal direction, with truly “diagonal” Gabor functions rarely being necessary for reconstruction of an STRF's main characteristics. As a toy example for the applicability of STRF and Gabor-STRF filters to auditory detection tasks, we use STRF filters as features in an automatic event detection task and compare them to idealized Gabor filters and mel-frequency cepstral coefficients (MFCCs). STRFs classify a set of six everyday sounds with an accuracy similar to reference Gabor features (94% recognition rate). Spectro-temporal STRF and Gabor features outperform reference spectral MFCCs in quiet and in low noise conditions (down to 0 dB signal to noise ratio).
Collapse
Affiliation(s)
- Jörg-Hendrik Bach
- Medizinische Physik, Universität OldenburgOldenburg, Germany
- Cluster of Excellence Hearing4all, Universität OldenburgOldenburg, Germany
| | - Birger Kollmeier
- Medizinische Physik, Universität OldenburgOldenburg, Germany
- Cluster of Excellence Hearing4all, Universität OldenburgOldenburg, Germany
| | - Jörn Anemüller
- Medizinische Physik, Universität OldenburgOldenburg, Germany
- Cluster of Excellence Hearing4all, Universität OldenburgOldenburg, Germany
- *Correspondence: Jörn Anemüller
| |
Collapse
|
18
|
Muhammad G, Alsulaiman M, Ali Z, Mesallam TA, Farahat M, Malki KH, Al-nasheri A, Bencherif MA. Voice pathology detection using interlaced derivative pattern on glottal source excitation. Biomed Signal Process Control 2017. [DOI: 10.1016/j.bspc.2016.08.002] [Citation(s) in RCA: 26] [Impact Index Per Article: 3.7] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/21/2022]
|
19
|
Stone MA, Canavan S. The near non-existence of "pure" energetic masking release for speech: Extension to spectro-temporal modulation and glimpsing. THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA 2016; 140:832. [PMID: 27586715 DOI: 10.1121/1.4960483] [Citation(s) in RCA: 22] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/06/2023]
Abstract
Stone and Moore [(2014). J. Acoust. Soc Am. 135, 1967-77], showed that the introduction of explicit temporal-only modulations to a speech masker, that otherwise produced a near-constant envelope at the output of each auditory filter, rarely resulted in improved intelligibility, except at a very low modulation rate. This represents a failure in "dip-listening" or "glimpsing" [Cooke (2006). J. Acoust. Soc. Am. 119, 1562-1573], a facility where listeners are presumed to benefit from the temporarily improved signal-to-noise ratio during the masker dips. The dips of Stone and Moore only varied temporally, so Stone and Moore's method was used here to investigate the effect of maskers with both spectral and temporal dips, a pattern more representative of real-world maskers. For sinusoidally shaped modulations, intelligibility improved only at very low modulation rates, below 2 Hz temporally and 0.14 ripples/auditory filter spectrally. Square-wave modulation at a rate of 4 Hz resulted in improved intelligibility when only one cycle of spectral modulation was present across the audio bandwidth. Compared to the spectro-temporal extent of dips present during real-world noisy speech, dips generated by the reported modulation patterns were very large, further supporting the notion that dip-listening reflects a release from modulation masking and not energetic masking.
Collapse
Affiliation(s)
- Michael A Stone
- Central Manchester University Hospitals NHS Foundation Trust, Manchester Academic Health Science Centre, Manchester, M13 9WL, United Kingdom
| | - Shanelle Canavan
- Manchester Centre for Audiology and Deafness, School of Health Sciences, University of Manchester, Manchester, M13 9PL, United Kingdom
| |
Collapse
|
20
|
A Review of Physical and Perceptual Feature Extraction Techniques for Speech, Music and Environmental Sounds. APPLIED SCIENCES-BASEL 2016. [DOI: 10.3390/app6050143] [Citation(s) in RCA: 46] [Impact Index Per Article: 5.8] [Reference Citation Analysis] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 11/16/2022]
|
21
|
Schädler MR, Warzybok A, Ewert SD, Kollmeier B. A simulation framework for auditory discrimination experiments: Revealing the importance of across-frequency processing in speech perception. THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA 2016; 139:2708. [PMID: 27250164 DOI: 10.1121/1.4948772] [Citation(s) in RCA: 16] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/05/2023]
Abstract
A framework for simulating auditory discrimination experiments, based on an approach from Schädler, Warzybok, Hochmuth, and Kollmeier [(2015). Int. J. Audiol. 54, 100-107] which was originally designed to predict speech recognition thresholds, is extended to also predict psychoacoustic thresholds. The proposed framework is used to assess the suitability of different auditory-inspired feature sets for a range of auditory discrimination experiments that included psychoacoustic as well as speech recognition experiments in noise. The considered experiments were 2 kHz tone-in-broadband-noise simultaneous masking depending on the tone length, spectral masking with simultaneously presented tone signals and narrow-band noise maskers, and German Matrix sentence test reception threshold in stationary and modulated noise. The employed feature sets included spectro-temporal Gabor filter bank features, Mel-frequency cepstral coefficients, logarithmically scaled Mel-spectrograms, and the internal representation of the Perception Model from Dau, Kollmeier, and Kohlrausch [(1997). J. Acoust. Soc. Am. 102(5), 2892-2905]. The proposed framework was successfully employed to simulate all experiments with a common parameter set and obtain objective thresholds with less assumptions compared to traditional modeling approaches. Depending on the feature set, the simulated reference-free thresholds were found to agree with-and hence to predict-empirical data from the literature. Across-frequency processing was found to be crucial to accurately model the lower speech reception threshold in modulated noise conditions than in stationary noise conditions.
Collapse
Affiliation(s)
- Marc René Schädler
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| | - Anna Warzybok
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| | - Stephan D Ewert
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| | - Birger Kollmeier
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| |
Collapse
|
22
|
Cortical Transformation of Spatial Processing for Solving the Cocktail Party Problem: A Computational Model(1,2,3). eNeuro 2016; 3:eN-NWR-0086-15. [PMID: 26866056 PMCID: PMC4745179 DOI: 10.1523/eneuro.0086-15.2015] [Citation(s) in RCA: 9] [Impact Index Per Article: 1.1] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/09/2015] [Revised: 12/16/2015] [Accepted: 12/18/2015] [Indexed: 12/04/2022] Open
Abstract
In multisource, “cocktail party” sound environments, human and animal auditory systems can use spatial cues to effectively separate and follow one source of sound over competing sources. In multisource, “cocktail party” sound environments, human and animal auditory systems can use spatial cues to effectively separate and follow one source of sound over competing sources. While mechanisms to extract spatial cues such as interaural time differences (ITDs) are well understood in precortical areas, how such information is reused and transformed in higher cortical regions to represent segregated sound sources is not clear. We present a computational model describing a hypothesized neural network that spans spatial cue detection areas and the cortex. This network is based on recent physiological findings that cortical neurons selectively encode target stimuli in the presence of competing maskers based on source locations (Maddox et al., 2012). We demonstrate that key features of cortical responses can be generated by the model network, which exploits spatial interactions between inputs via lateral inhibition, enabling the spatial separation of target and interfering sources while allowing monitoring of a broader acoustic space when there is no competition. We present the model network along with testable experimental paradigms as a starting point for understanding the transformation and organization of spatial information from midbrain to cortex. This network is then extended to suggest engineering solutions that may be useful for hearing-assistive devices in solving the cocktail party problem.
Collapse
|
23
|
Shen L, Zhao L, Hong B. Frequency-specific adaptation and its underlying circuit model in the auditory midbrain. Front Neural Circuits 2015; 9:55. [PMID: 26483641 PMCID: PMC4589587 DOI: 10.3389/fncir.2015.00055] [Citation(s) in RCA: 7] [Impact Index Per Article: 0.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/22/2015] [Accepted: 09/16/2015] [Indexed: 11/13/2022] Open
Abstract
Receptive fields of sensory neurons are considered to be dynamic and depend on the stimulus history. In the auditory system, evidence of dynamic frequency-receptive fields has been found following stimulus-specific adaptation (SSA). However, the underlying mechanism and circuitry of SSA have not been fully elucidated. Here, we studied how frequency-receptive fields of neurons in rat inferior colliculus (IC) changed when exposed to a biased tone sequence. Pure tone with one specific frequency (adaptor) was presented markedly more often than others. The adapted tuning was compared with the original tuning measured with an unbiased sequence. We found inhomogeneous changes in frequency tuning in IC, exhibiting a center-surround pattern with respect to the neuron's best frequency. Central adaptors elicited strong suppressive and repulsive changes while flank adaptors induced facilitative and attractive changes. Moreover, we proposed a two-layer model of the underlying network, which not only reproduced the adaptive changes in the receptive fields but also predicted novelty responses to oddball sequences. These results suggest that frequency-specific adaptation in auditory midbrain can be accounted for by an adapted frequency channel and its lateral spreading of adaptation, which shed light on the organization of the underlying circuitry.
Collapse
Affiliation(s)
- Li Shen
- Department of Biomedical Engineering, School of Medicine, Tsinghua University Beijing, China
| | - Lingyun Zhao
- Department of Biomedical Engineering, School of Medicine, Tsinghua University Beijing, China
| | - Bo Hong
- Department of Biomedical Engineering, School of Medicine, Tsinghua University Beijing, China
| |
Collapse
|
24
|
Mahmood A, Alsulaiman M, Muhammad G, Akram S. Artificially intelligent recognition of Arabic speaker using voice print-based local features. J EXP THEOR ARTIF IN 2015. [DOI: 10.1080/0952813x.2015.1055827] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/23/2022]
|
25
|
Schädler MR, Kollmeier B. Separable spectro-temporal Gabor filter bank features: Reducing the complexity of robust features for automatic speech recognition. THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA 2015; 137:2047-2059. [PMID: 25920855 DOI: 10.1121/1.4916618] [Citation(s) in RCA: 16] [Impact Index Per Article: 1.8] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/04/2023]
Abstract
To test if simultaneous spectral and temporal processing is required to extract robust features for automatic speech recognition (ASR), the robust spectro-temporal two-dimensional-Gabor filter bank (GBFB) front-end from Schädler, Meyer, and Kollmeier [J. Acoust. Soc. Am. 131, 4134-4151 (2012)] was de-composed into a spectral one-dimensional-Gabor filter bank and a temporal one-dimensional-Gabor filter bank. A feature set that is extracted with these separate spectral and temporal modulation filter banks was introduced, the separate Gabor filter bank (SGBFB) features, and evaluated on the CHiME (Computational Hearing in Multisource Environments) keywords-in-noise recognition task. From the perspective of robust ASR, the results showed that spectral and temporal processing can be performed independently and are not required to interact with each other. Using SGBFB features permitted the signal-to-noise ratio (SNR) to be lowered by 1.2 dB while still performing as well as the GBFB-based reference system, which corresponds to a relative improvement of the word error rate by 12.8%. Additionally, the real time factor of the spectro-temporal processing could be reduced by more than an order of magnitude. Compared to human listeners, the SNR needed to be 13 dB higher when using Mel-frequency cepstral coefficient features, 11 dB higher when using GBFB features, and 9 dB higher when using SGBFB features to achieve the same recognition performance.
Collapse
Affiliation(s)
- Marc René Schädler
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| | - Birger Kollmeier
- Medizinische Physik and Cluster of Excellence Hearing4all, Universität Oldenburg, D-26111 Oldenburg, Germany
| |
Collapse
|
26
|
Lindeberg T, Friberg A. Idealized computational models for auditory receptive fields. PLoS One 2015; 10:e0119032. [PMID: 25822973 PMCID: PMC4379182 DOI: 10.1371/journal.pone.0119032] [Citation(s) in RCA: 11] [Impact Index Per Article: 1.2] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/04/2014] [Accepted: 01/24/2015] [Indexed: 11/19/2022] Open
Abstract
We present a theory by which idealized models of auditory receptive fields can be derived in a principled axiomatic manner, from a set of structural properties to (i) enable invariance of receptive field responses under natural sound transformations and (ii) ensure internal consistency between spectro-temporal receptive fields at different temporal and spectral scales. For defining a time-frequency transformation of a purely temporal sound signal, it is shown that the framework allows for a new way of deriving the Gabor and Gammatone filters as well as a novel family of generalized Gammatone filters, with additional degrees of freedom to obtain different trade-offs between the spectral selectivity and the temporal delay of time-causal temporal window functions. When applied to the definition of a second-layer of receptive fields from a spectrogram, it is shown that the framework leads to two canonical families of spectro-temporal receptive fields, in terms of spectro-temporal derivatives of either spectro-temporal Gaussian kernels for non-causal time or a cascade of time-causal first-order integrators over the temporal domain and a Gaussian filter over the logspectral domain. For each filter family, the spectro-temporal receptive fields can be either separable over the time-frequency domain or be adapted to local glissando transformations that represent variations in logarithmic frequencies over time. Within each domain of either non-causal or time-causal time, these receptive field families are derived by uniqueness from the assumptions. It is demonstrated how the presented framework allows for computation of basic auditory features for audio processing and that it leads to predictions about auditory receptive fields with good qualitative similarity to biological receptive fields measured in the inferior colliculus (ICC) and primary auditory cortex (A1) of mammals.
Collapse
Affiliation(s)
- Tony Lindeberg
- Department of Computational Biology, School of Computer Science and Communication, KTH Royal Institute of Technology, Stockholm, Sweden
| | - Anders Friberg
- Department of Speech, Music and Hearing, School of Computer Science and Communication, KTH Royal Institute of Technology, Stockholm, Sweden
| |
Collapse
|
27
|
Koka K, Tollin DJ. Linear coding of complex sound spectra by discharge rate in neurons of the medial nucleus of the trapezoid body (MNTB) and its inputs. Front Neural Circuits 2014; 8:144. [PMID: 25565971 PMCID: PMC4267272 DOI: 10.3389/fncir.2014.00144] [Citation(s) in RCA: 13] [Impact Index Per Article: 1.3] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/15/2014] [Accepted: 11/25/2014] [Indexed: 11/25/2022] Open
Abstract
The interaural level difference (ILD) cue to sound location is first encoded in the lateral superior olive (LSO). ILD sensitivity results because the LSO receives excitatory input from the ipsilateral cochlear nucleus and inhibitory input indirectly from the contralateral cochlear nucleus via glycinergic neurons of the ipsilateral medial nucleus of the trapezoid body (MNTB). It is hypothesized that in order for LSO neurons to encode ILDs, the sound spectra at both ears must be accurately encoded via spike rate by their afferents. This spectral-coding hypothesis has not been directly tested in MNTB, likely because MNTB neurons have been mostly described and studied recently in regards to their abilities to encode temporal aspects of sounds, not spectral. Here, we test the hypothesis that MNTB neurons and their inputs from the cochlear nucleus and auditory nerve code sound spectra via discharge rate. The Random Spectral Shape (RSS) method was used to estimate how the levels of 100-ms duration spectrally stationary stimuli were weighted, both linearly and non-linearly, across a wide band of frequencies. In general, MNTB neurons, and their globular bushy cell inputs, were found to be well-modeled by a linear weighting of spectra demonstrating that the pathways through the MNTB can accurately encode sound spectra including those resulting from the acoustical cues to sound location provided by head-related directional transfer functions (DTFs). Together with the anatomical and biophysical specializations for timing in the MNTB-LSO complex, these mechanisms may allow ILDs to be computed for complex stimuli with rapid spectrotemporally-modulated envelopes such as speech and animal vocalizations and moving sound sources.
Collapse
Affiliation(s)
- Kanthaiah Koka
- Department of Physiology and Biophysics, University of Colorado School of Medicine Aurora, CO, USA
| | - Daniel J Tollin
- Department of Physiology and Biophysics, University of Colorado School of Medicine Aurora, CO, USA
| |
Collapse
|
28
|
Escabí MA, Read HL, Viventi J, Kim DH, Higgins NC, Storace DA, Liu ASK, Gifford AM, Burke JF, Campisi M, Kim YS, Avrin AE, Spiegel Jan VD, Huang Y, Li M, Wu J, Rogers JA, Litt B, Cohen YE. A high-density, high-channel count, multiplexed μECoG array for auditory-cortex recordings. J Neurophysiol 2014; 112:1566-83. [PMID: 24920021 DOI: 10.1152/jn.00179.2013] [Citation(s) in RCA: 72] [Impact Index Per Article: 7.2] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
Our understanding of the large-scale population dynamics of neural activity is limited, in part, by our inability to record simultaneously from large regions of the cortex. Here, we validated the use of a large-scale active microelectrode array that simultaneously records 196 multiplexed micro-electrocortigraphical (μECoG) signals from the cortical surface at a very high density (1,600 electrodes/cm(2)). We compared μECoG measurements in auditory cortex using a custom "active" electrode array to those recorded using a conventional "passive" μECoG array. Both of these array responses were also compared with data recorded via intrinsic optical imaging, which is a standard methodology for recording sound-evoked cortical activity. Custom active μECoG arrays generated more veridical representations of the tonotopic organization of the auditory cortex than current commercially available passive μECoG arrays. Furthermore, the cortical representation could be measured efficiently with the active arrays, requiring as little as 13.5 s of neural data acquisition. Next, we generated spectrotemporal receptive fields from the recorded neural activity on the active μECoG array and identified functional organizational principles comparable to those observed using intrinsic metabolic imaging and single-neuron recordings. This new electrode array technology has the potential for large-scale, temporally precise monitoring and mapping of the cortex, without the use of invasive penetrating electrodes.
Collapse
Affiliation(s)
- Monty A Escabí
- Department of Psychology, University of Connecticut, Storrs, Connecticut; Department of Biomedical Engineering, University of Connecticut, Storrs, Connecticut; Department of Electrical Engineering, University of Connecticut, Storrs, Connecticut
| | - Heather L Read
- Department of Psychology, University of Connecticut, Storrs, Connecticut; Department of Biomedical Engineering, University of Connecticut, Storrs, Connecticut
| | - Jonathan Viventi
- Center for Neural Science, New York University, New York, New York; Department of Electrical and Computer Engineering, Polytechnic Institute of New York University, Brooklyn, New York
| | - Dae-Hyeong Kim
- Center for Nanoparticle Research of Institute for Basic Science, School of Chemical and Biological Engineering, Seoul National University, Seoul, Republic of Korea
| | - Nathan C Higgins
- Department of Psychology, University of Connecticut, Storrs, Connecticut
| | - Douglas A Storace
- Department of Psychology, University of Connecticut, Storrs, Connecticut
| | - Andrew S K Liu
- Bioengineering Graduate Group, University of Pennsylvania, Philadelphia, Pennsylvania
| | - Adam M Gifford
- Neuroscience Graduate Group, University of Pennsylvania, Philadelphia, Pennsylvania
| | - John F Burke
- Neuroscience Graduate Group, University of Pennsylvania, Philadelphia, Pennsylvania
| | - Matthew Campisi
- Department of Electrical and Computer Engineering, Polytechnic Institute of New York University, Brooklyn, New York
| | - Yun-Soung Kim
- Department of Materials Science and Engineering, Beckman Institute for Advanced Science and Technology and Frederick Seitz Materials Research Laboratory, University of Illinois at Urbana-Champaign, Urbana, Illinois
| | - Andrew E Avrin
- Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, Pennsylvania
| | - Van der Spiegel Jan
- Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, Pennsylvania
| | - Yonggang Huang
- Departments of Mechanical Engineering and Civil and Environmental Engineering, Northwestern University, Evanston, Illinois
| | - Ming Li
- State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University of Technology, Dalian, China
| | - Jian Wu
- Department of Engineering Mechanics, Tsinghua University, Beijing, China
| | - John A Rogers
- Department of Materials Science and Engineering, Beckman Institute for Advanced Science and Technology and Frederick Seitz Materials Research Laboratory, University of Illinois at Urbana-Champaign, Urbana, Illinois
| | - Brian Litt
- Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania; Department of Neurology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania
| | - Yale E Cohen
- Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania; Department of Neuroscience, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; and Department of Otorhinolaryngology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania
| |
Collapse
|
29
|
Młynarski W. Efficient coding of spectrotemporal binaural sounds leads to emergence of the auditory space representation. Front Comput Neurosci 2014; 8:26. [PMID: 24639644 PMCID: PMC3945936 DOI: 10.3389/fncom.2014.00026] [Citation(s) in RCA: 4] [Impact Index Per Article: 0.4] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/04/2013] [Accepted: 02/19/2014] [Indexed: 11/23/2022] Open
Abstract
To date a number of studies have shown that receptive field shapes of early sensory neurons can be reproduced by optimizing coding efficiency of natural stimulus ensembles. A still unresolved question is whether the efficient coding hypothesis explains formation of neurons which explicitly represent environmental features of different functional importance. This paper proposes that the spatial selectivity of higher auditory neurons emerges as a direct consequence of learning efficient codes for natural binaural sounds. Firstly, it is demonstrated that a linear efficient coding transform—Independent Component Analysis (ICA) trained on spectrograms of naturalistic simulated binaural sounds extracts spatial information present in the signal. A simple hierarchical ICA extension allowing for decoding of sound position is proposed. Furthermore, it is shown that units revealing spatial selectivity can be learned from a binaural recording of a natural auditory scene. In both cases a relatively small subpopulation of learned spectrogram features suffices to perform accurate sound localization. Representation of the auditory space is therefore learned in a purely unsupervised way by maximizing the coding efficiency and without any task-specific constraints. This results imply that efficient coding is a useful strategy for learning structures which allow for making behaviorally vital inferences about the environment.
Collapse
Affiliation(s)
- Wiktor Młynarski
- Max-Planck Institute for Mathematics in the Sciences Leipzig, Germany
| |
Collapse
|
30
|
Bandyopadhyay S, Young ED. Nonlinear temporal receptive fields of neurons in the dorsal cochlear nucleus. J Neurophysiol 2013; 110:2414-25. [PMID: 23986561 DOI: 10.1152/jn.00278.2013] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
Studies of the dorsal cochlear nucleus (DCN) have focused on spectral processing because of the complex spectral receptive fields of the DCN. However, temporal fluctuations in natural signals convey important information, including information about moving sound sources or movements of the external ear in animals like cats. Here, we investigate the temporal filtering properties of DCN principal neurons through the use of temporal weighting functions that allow flexible analysis of nonlinearities and time variation in temporal response properties. First-order temporal receptive fields derived from the neurons are sufficient to characterize their response properties to low-contrast (3-dB standard deviation) stimuli. Larger contrasts require the second-order terms. Allowing temporal variation of the parameters of the first-order model or adding a component representing refractoriness improves predictions by the model by relatively small amounts. The importance of second-order components of the model is shown through simulations of nonlinear envelope synchronization behavior across sound level. The temporal model can be combined with a spectral model to predict tuning to the speed and direction of moving sounds.
Collapse
|
31
|
Yu JJ, Young ED. Frequency response areas in the inferior colliculus: nonlinearity and binaural interaction. Front Neural Circuits 2013; 7:90. [PMID: 23675323 PMCID: PMC3650518 DOI: 10.3389/fncir.2013.00090] [Citation(s) in RCA: 10] [Impact Index Per Article: 0.9] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 01/28/2013] [Accepted: 04/22/2013] [Indexed: 11/13/2022] Open
Abstract
The tuning, binaural properties, and encoding characteristics of neurons in the central nucleus of the inferior colliculus (CNIC) were investigated to shed light on nonlinearities in the responses of these neurons. Results were analyzed for three types of neurons (I, O, and V) in the CNIC of decerebrate cats. Rate responses to binaural stimuli were characterized using a 1st- plus 2nd-order spectral integration model. Parameters of the model were derived using broadband stimuli with random spectral shapes (RSS). This method revealed four characteristics of CNIC neurons: (1) Tuning curves derived from broadband stimuli have fixed (i. e., level tolerant) bandwidths across a 50-60 dB range of sound levels; (2) 1st-order contralateral weights (particularly for type I and O neurons) were usually larger in magnitude than corresponding ipsilateral weights; (3) contralateral weights were more important than ipsilateral weights when using the model to predict responses to untrained noise stimuli; and (4) 2nd-order weight functions demonstrate frequency selectivity different from that of 1st-order weight functions. Furthermore, while the inclusion of 2nd-order terms in the model usually improved response predictions related to untrained RSS stimuli, they had limited impact on predictions related to other forms of filtered broadband noise [e. g., virtual-space stimuli (VS)]. The accuracy of the predictions varied considerably by response type. Predictions were most accurate for I neurons, and less accurate for O and V neurons, except at the lowest stimulus levels. These differences in prediction performance support the idea that type I, O, and V neurons encode different aspects of the stimulus: while type I neurons are most capable of producing linear representations of spectral shape, type O and V neurons may encode spectral features or temporal stimulus properties in a manner not easily explained with the low-order model. Supported by NIH grant DC00115.
Collapse
Affiliation(s)
| | - Eric D. Young
- Center for Hearing and Balance, Department of Biomedical Engineering, Johns Hopkins UniversityBaltimore, MD, USA
| |
Collapse
|
32
|
Chang TR, Chiu TW, Sun X, Poon PWF. Modeling complex responses of FM-sensitive cells in the auditory midbrain using a committee machine. Brain Res 2013; 1536:44-52. [PMID: 23665390 DOI: 10.1016/j.brainres.2013.04.058] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.1] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/31/2012] [Revised: 04/30/2013] [Accepted: 04/30/2013] [Indexed: 11/26/2022]
Abstract
Frequency modulation (FM) is an important building block of complex sounds that include speech signals. Exploring the neural mechanisms of FM coding with computer modeling could help understand how speech sounds are processed in the brain. Here, we modeled the single unit responses of auditory neurons recorded from the midbrain of anesthetized rats. These neurons displayed spectral temporal receptive fields (STRFs) that had multiple-trigger features, and were more complex than those with single-trigger features. Their responses have not been modeled satisfactorily with simple artificial neural networks, unlike neurons with simple-trigger features. To improve model performance, here we tested an approach with the committee machine. For a given neuron, the peri-stimulus time histogram (PSTH) was first generated in response to a repeated random FM tone, and peaks in the PSTH were segregated into groups based on the similarity of their pre-spike FM trigger features. Each group was then modeled using an artificial neural network with simple architecture, and, when necessary, by increasing the number of neurons in the hidden layer. After initial training, the artificial neural networks with their optimized weighting coefficients were pooled into a committee machine for training. Finally, the model performance was tested by prediction of the response of the same cell to a novel FM tone. The results showed improvement over simple artificial neural networks, supporting that trigger-feature-based modeling can be extended to cells with complex responses. This article is part of a Special Issue entitled Neural Coding 2012. This article is part of a Special Issue entitled Neural Coding 2012.
Collapse
Affiliation(s)
- T R Chang
- Department of Computer Science and Information Engineering, Southern Taiwan University of Science and Technology, Tainan, Taiwan.
| | | | | | | |
Collapse
|
33
|
Chen C, Rodriguez FC, Read HL, Escabí MA. Spectrotemporal sound preferences of neighboring inferior colliculus neurons: implications for local circuitry and processing. Front Neural Circuits 2012; 6:62. [PMID: 23060750 PMCID: PMC3461703 DOI: 10.3389/fncir.2012.00062] [Citation(s) in RCA: 26] [Impact Index Per Article: 2.2] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/30/2012] [Accepted: 08/19/2012] [Indexed: 11/13/2022] Open
Abstract
How do local circuits in the inferior colliculus (IC) process and transform spectral and temporal sound information? Using a four-tetrode array we examined the functional properties of the IC and metrics of its micro circuitry by recording neural activity from neighboring single neurons in the cat. Spectral and temporal response preferences were compared for neurons found on the same and adjacent tetrodes (ATs), as well as across distant recording sites. We found that neighboring neurons had similar preferences while neurons recorded across distant sites were less similar. Best frequency (BF) was the most correlated parameter between neighboring neurons and BF differences exhibited unique clustering at ~0.3 octave intervals, indicative of the frequency band lamina. Other spectral and temporal parameters of the receptive fields were more similar for neighboring neurons than for those at distant sites and the receptive field similarity was larger for neurons with small differences in BF. Furthermore, correlated firing was stronger for neighboring neuron pairs and increased with proximity and decreasing BF difference. Thus, although response selectivities are quite diverse in the IC, spectral, and temporal preference within a local microcircuit are functionally quite similar. This suggests a scheme where local circuits are organized into zones that are specialized for processing distinct spectrotemporal cues.
Collapse
Affiliation(s)
- Chen Chen
- Department of Electrical and Computer Engineering, University of Connecticut Storrs, CT, USA
| | | | | | | |
Collapse
|
34
|
Precise feature based time scales and frequency decorrelation lead to a sparse auditory code. J Neurosci 2012; 32:8454-68. [PMID: 22723685 DOI: 10.1523/jneurosci.6506-11.2012] [Citation(s) in RCA: 30] [Impact Index Per Article: 2.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/21/2022] Open
Abstract
Sparse redundancy reducing codes have been proposed as efficient strategies for representing sensory stimuli. A prevailing hypothesis suggests that sensory representations shift from dense redundant codes in the periphery to selective sparse codes in cortex. We propose an alternative framework where sparseness and redundancy depend on sensory integration time scales and demonstrate that the central nucleus of the inferior colliculus (ICC) of cats encodes sound features by precise sparse spike trains. Direct comparisons with auditory cortical neurons demonstrate that ICC responses were sparse and uncorrelated as long as the spike train time scales were matched to the sensory integration time scales relevant to ICC neurons. Intriguingly, correlated spiking in the ICC was substantially lower than predicted by linear or nonlinear models and strictly observed for neurons with best frequencies within a "critical band," the hallmark of perceptual frequency resolution in mammals. This is consistent with a sparse asynchronous code throughout much of the ICC and a complementary correlation code within a critical band that may allow grouping of perceptually relevant cues.
Collapse
|
35
|
Carlson NL, Ming VL, DeWeese MR. Sparse codes for speech predict spectrotemporal receptive fields in the inferior colliculus. PLoS Comput Biol 2012; 8:e1002594. [PMID: 22807665 PMCID: PMC3395612 DOI: 10.1371/journal.pcbi.1002594] [Citation(s) in RCA: 41] [Impact Index Per Article: 3.4] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/16/2011] [Accepted: 05/18/2012] [Indexed: 12/03/2022] Open
Abstract
We have developed a sparse mathematical representation of speech that minimizes the number of active model neurons needed to represent typical speech sounds. The model learns several well-known acoustic features of speech such as harmonic stacks, formants, onsets and terminations, but we also find more exotic structures in the spectrogram representation of sound such as localized checkerboard patterns and frequency-modulated excitatory subregions flanked by suppressive sidebands. Moreover, several of these novel features resemble neuronal receptive fields reported in the Inferior Colliculus (IC), as well as auditory thalamus and cortex, and our model neurons exhibit the same tradeoff in spectrotemporal resolution as has been observed in IC. To our knowledge, this is the first demonstration that receptive fields of neurons in the ascending mammalian auditory pathway beyond the auditory nerve can be predicted based on coding principles and the statistical properties of recorded sounds.
Collapse
Affiliation(s)
- Nicole L. Carlson
- Redwood Center for Theoretical Neuroscience, University of California, Berkeley, California, United States of America
- Department of Physics, University of California, Berkeley, California, United States of America
| | - Vivienne L. Ming
- Redwood Center for Theoretical Neuroscience, University of California, Berkeley, California, United States of America
| | - Michael Robert DeWeese
- Redwood Center for Theoretical Neuroscience, University of California, Berkeley, California, United States of America
- Department of Physics, University of California, Berkeley, California, United States of America
- Helen Wills Neuroscience Institute, University of California, Berkeley, California, United States of America
| |
Collapse
|
36
|
Schädler MR, Meyer BT, Kollmeier B. Spectro-temporal modulation subspace-spanning filter bank features for robust automatic speech recognition. THE JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA 2012; 131:4134-4151. [PMID: 22559385 DOI: 10.1121/1.3699200] [Citation(s) in RCA: 30] [Impact Index Per Article: 2.5] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 05/31/2023]
Abstract
In an attempt to increase the robustness of automatic speech recognition (ASR) systems, a feature extraction scheme is proposed that takes spectro-temporal modulation frequencies (MF) into account. This physiologically inspired approach uses a two-dimensional filter bank based on Gabor filters, which limits the redundant information between feature components, and also results in physically interpretable features. Robustness against extrinsic variation (different types of additive noise) and intrinsic variability (arising from changes in speaking rate, effort, and style) is quantified in a series of recognition experiments. The results are compared to reference ASR systems using Mel-frequency cepstral coefficients (MFCCs), MFCCs with cepstral mean subtraction (CMS) and RASTA-PLP features, respectively. Gabor features are shown to be more robust against extrinsic variation than the baseline systems without CMS, with relative improvements of 28% and 16% for two training conditions (using only clean training samples or a mixture of noisy and clean utterances, respectively). When used in a state-of-the-art system, improvements of 14% are observed when spectro-temporal features are concatenated with MFCCs, indicating the complementarity of those feature types. An analysis of the importance of specific MF shows that temporal MF up to 25 Hz and spectral MF up to 0.25 cycles/channel are beneficial for ASR.
Collapse
Affiliation(s)
- Marc René Schädler
- Medizinische Physik, Carl-von-Ossietzky Universität Oldenburg, D-26111 Oldenburg, Germany.
| | | | | |
Collapse
|
37
|
Selectivity for spectral motion as a neural computation for encoding natural communication signals in bat inferior colliculus. J Neurosci 2012; 31:16529-40. [PMID: 22090479 DOI: 10.1523/jneurosci.1306-11.2011] [Citation(s) in RCA: 29] [Impact Index Per Article: 2.4] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/21/2022] Open
Abstract
This study examines the neural computations performed by neurons in the auditory system to be selective for the direction and velocity of signals sweeping upward or downward in frequency, termed spectral motion. We show that neurons in the auditory midbrain of Mexican free-tailed bats encode multiple spectrotemporal features of natural communication sounds. These features to which each neuron is tuned are nonlinearly combined to produce selectivity for spectral motion cues present in their conspecific calls, such as direction and velocity. We find that the neural computations resulting in selectivity for spectral motion are analogous to models of motion selectivity studied in vision. Our analysis revealed that auditory neurons in the inferior colliculus (IC) are avoiding spectrotemporal modulations that are redundant across different bat communication signals and are specifically tuned for modulations that distinguish each call from another by their frequency-modulated direction and velocity, suggesting that spectral motion is the neural computation through which IC neurons are encoding specific features of conspecific vocalizations.
Collapse
|
38
|
Effect of instantaneous frequency glides on interaural time difference processing by auditory coincidence detectors. Proc Natl Acad Sci U S A 2011; 108:18138-43. [PMID: 22006305 DOI: 10.1073/pnas.1108921108] [Citation(s) in RCA: 16] [Impact Index Per Article: 1.2] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/18/2022] Open
Abstract
Detecting interaural time difference (ITD) is crucial for sound localization. The temporal accuracy required to detect ITD, and how ITD is initially encoded, continue to puzzle scientists. A fundamental question is whether the monaural inputs to the binaural ITD detectors differ only in their timing, when temporal and spectral tunings are largely inseparable in the auditory pathway. Here, we investigate the spectrotemporal selectivity of the monaural inputs to ITD detector neurons of the owl. We found that these inputs are selective for instantaneous frequency glides. Modeling shows that ITD tuning depends strongly on whether the monaural inputs are spectrotemporally matched, an effect that may generalize to mammals. We compare the spectrotemporal selectivity of monaural inputs of ITD detector neurons in vivo, demonstrating that their selectivity matches. Finally, we show that this refinement can develop through spike timing-dependent plasticity. Our findings raise the unexplored issue of time-dependent frequency tuning in auditory coincidence detectors and offer a unifying perspective.
Collapse
|
39
|
Ding N, Simon JZ. Neural coding of continuous speech in auditory cortex during monaural and dichotic listening. J Neurophysiol 2011; 107:78-89. [PMID: 21975452 DOI: 10.1152/jn.00297.2011] [Citation(s) in RCA: 273] [Impact Index Per Article: 21.0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
The cortical representation of the acoustic features of continuous speech is the foundation of speech perception. In this study, noninvasive magnetoencephalography (MEG) recordings are obtained from human subjects actively listening to spoken narratives, in both simple and cocktail party-like auditory scenes. By modeling how acoustic features of speech are encoded in ongoing MEG activity as a spectrotemporal response function, we demonstrate that the slow temporal modulations of speech in a broad spectral region are represented bilaterally in auditory cortex by a phase-locked temporal code. For speech presented monaurally to either ear, this phase-locked response is always more faithful in the right hemisphere, but with a shorter latency in the hemisphere contralateral to the stimulated ear. When different spoken narratives are presented to each ear simultaneously (dichotic listening), the resulting cortical neural activity precisely encodes the acoustic features of both of the spoken narratives, but slightly weakened and delayed compared with the monaural response. Critically, the early sensory response to the attended speech is considerably stronger than that to the unattended speech, demonstrating top-down attentional gain control. This attentional gain is substantial even during the subjects' very first exposure to the speech mixture and therefore largely independent of knowledge of the speech content. Together, these findings characterize how the spectrotemporal features of speech are encoded in human auditory cortex and establish a single-trial-based paradigm to study the neural basis underlying the cocktail party phenomenon.
Collapse
Affiliation(s)
- Nai Ding
- Univ. of Maryland, College Park, MD 20742, USA
| | | |
Collapse
|
40
|
Chang TR, Chiu TW, Sun X, Poon PWF. Modeling frequency modulated responses of midbrain auditory neurons based on trigger features and artificial neural networks. Brain Res 2011; 1434:90-101. [PMID: 22035565 DOI: 10.1016/j.brainres.2011.09.042] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.2] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/14/2011] [Revised: 09/20/2011] [Accepted: 09/21/2011] [Indexed: 11/25/2022]
Abstract
Frequency modulation (FM) is an important building block of communication signals for animals and human. Attempts to predict the response of central neurons to FM sounds have not been very successful, though achieving successful results could bring insights regarding the underlying neural mechanisms. Here we proposed a new method to predict responses of FM-sensitive neurons in the auditory midbrain. First we recorded single unit responses in anesthetized rats using a random FM tone to construct their spectro-temporal receptive fields (STRFs). Training of neurons in the artificial neural network to respond to a second random FM tone was based on the temporal information derived from the STRF. Specifically, the time window covered by the presumed trigger feature and its delay time to spike occurrence were used to train a finite impulse response neural network (FIRNN) to respond to this random FM. Finally we tested the model performance in predicting the response to another similar FM stimuli (third random FM tone). We found good performance in predicting the time of responses if not also the response magnitudes. Furthermore, the weighting function of the FIRNN showed temporal 'bumps' suggesting temporal integration of synaptic inputs from different frequency laminae. This article is part of a Special Issue entitled: Neural Coding.
Collapse
Affiliation(s)
- T R Chang
- Dept. of Computer Sciences and Information Engineering, Southern Taiwan University, Tainan, Taiwan.
| | | | | | | |
Collapse
|
41
|
Neural modulation tuning characteristics scale to efficiently encode natural sound statistics. J Neurosci 2010; 30:15969-80. [PMID: 21106835 DOI: 10.1523/jneurosci.0966-10.2010] [Citation(s) in RCA: 55] [Impact Index Per Article: 3.9] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/21/2022] Open
Abstract
The efficient-coding hypothesis asserts that neural and perceptual sensitivity evolved to faithfully represent biologically relevant sensory signals. Here we characterized the spectrotemporal modulation statistics of several natural sound ensembles and examined how neurons encode these statistics in the central nucleus of the inferior colliculus (CNIC) of cats. We report that modulation-tuning in the CNIC is matched to equalize the modulation power of natural sounds. Specifically, natural sounds exhibited a tradeoff between spectral and temporal modulations, which manifests as 1/f modulation power spectrum (MPS). Neural tuning was highly overlapped with the natural sound MPS and neurons approximated proportional resolution filters where modulation bandwidths scaled with characteristic modulation frequencies, a behavior previously described in human psychoacoustics. We demonstrate that this neural scaling opposes the 1/f scaling of natural sounds and enhances the natural sound representation by equalizing their MPS. Modulation tuning in the CNIC may thus have evolved to represent natural sound modulations in a manner consistent with efficiency principles and the resulting characteristics likely underlie perceptual resolution.
Collapse
|
42
|
O'Connor KN, Yin P, Petkov CI, Sutter ML. Complex spectral interactions encoded by auditory cortical neurons: relationship between bandwidth and pattern. Front Syst Neurosci 2010; 4:145. [PMID: 21152347 PMCID: PMC2998047 DOI: 10.3389/fnsys.2010.00145] [Citation(s) in RCA: 37] [Impact Index Per Article: 2.6] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/26/2010] [Accepted: 09/09/2010] [Indexed: 11/13/2022] Open
Abstract
The focus of most research on auditory cortical neurons has concerned the effects of rather simple stimuli, such as pure tones or broad-band noise, or the modulation of a single acoustic parameter. Extending these findings to feature coding in more complex stimuli such as natural sounds may be difficult, however. Generalizing results from the simple to more complex case may be complicated by non-linear interactions occurring between multiple, simultaneously varying acoustic parameters in complex sounds. To examine this issue in the frequency domain, we performed a parametric study of the effects of two global features, spectral pattern (here ripple frequency) and bandwidth, on primary auditory (A1) neurons in awake macaques. Most neurons were tuned for one or both variables and most also displayed an interaction between bandwidth and pattern implying that their effects were conditional or interdependent. A spectral linear filter model was able to qualitatively reproduce the basic effects and interactions, indicating that a simple neural mechanism may be able to account for these interdependencies. Our results suggest that the behavior of most A1 neurons is likely to depend on multiple parameters, and so most are unlikely to respond independently or invariantly to specific acoustic features.
Collapse
Affiliation(s)
- Kevin N O'Connor
- Center for Neuroscience, University of California Davis Davis, CA, USA
| | | | | | | |
Collapse
|
43
|
Amin N, Gill P, Theunissen FE. Role of the zebra finch auditory thalamus in generating complex representations for natural sounds. J Neurophysiol 2010; 104:784-98. [PMID: 20554842 DOI: 10.1152/jn.00128.2010] [Citation(s) in RCA: 21] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
We estimated the spectrotemporal receptive fields of neurons in the songbird auditory thalamus, nucleus ovoidalis, and compared the neural representation of complex sounds in the auditory thalamus to those found in the upstream auditory midbrain nucleus, mesencephalicus lateralis dorsalis (MLd), and the downstream auditory pallial region, field L. Our data refute the idea that the primary sensory thalamus acts as a simple, relay nucleus: we find that the auditory thalamic receptive fields obtained in response to song are more complex than the ones found in the midbrain. Moreover, we find that linear tuning diversity and complexity in ovoidalis (Ov) are closer to those found in field L than in MLd. We also find prevalent tuning to intermediate spectral and temporal modulations, a feature that is unique to Ov. Thus even a feed-forward model of the sensory processing chain, where neural responses in the sensory thalamus reveals intermediate response properties between those in the sensory periphery and those in the primary sensory cortex, is inadequate in describing the tuning found in Ov. Based on these results, we believe that the auditory thalamic circuitry plays an important role in generating novel complex representations for specific features found in natural sounds.
Collapse
Affiliation(s)
- Noopur Amin
- Helen Wills Neuroscience Institute, University of California, Berkeley, California 94720-1650, USA
| | | | | |
Collapse
|
44
|
Rodríguez FA, Read HL, Escabí MA. Spectral and temporal modulation tradeoff in the inferior colliculus. J Neurophysiol 2009; 103:887-903. [PMID: 20018831 DOI: 10.1152/jn.00813.2009] [Citation(s) in RCA: 63] [Impact Index Per Article: 4.2] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
The cochlea encodes sounds through frequency-selective channels that exhibit low-pass modulation sensitivity. Unlike the cochlea, neurons in the auditory midbrain are tuned for spectral and temporal modulations found in natural sounds, yet the role of this transformation is not known. We report a distinct tradeoff in modulation sensitivity and tuning that is topographically ordered within the central nucleus of the inferior colliculus (CNIC). Spectrotemporal receptive fields (STRFs) were obtained with 16-channel electrodes inserted orthogonal to the isofrequency lamina. Surprisingly, temporal and spectral characteristics exhibited an opposing relationship along the tonotopic axis. For low best frequencies (BFs), units were selective for fast temporal and broad spectral modulations. A systematic progression was observed toward slower temporal and finer spectral modulation sensitivity at high BF. This tradeoff was strongly reflected in the arrangement of excitation and inhibition and, consequently, in the modulation tuning characteristics. Comparisons with auditory nerve fibers show that these trends oppose the pattern imposed by the peripheral filters. These results suggest that spectrotemporal preferences are reordered within the tonotopic axis of the CNIC. This topographic organization has profound implications for the coding of spectrotemporal features in natural sounds and could underlie a number of perceptual phenomena.
Collapse
|
45
|
Abstract
Auditory perception depends on the coding and organization of the information-bearing acoustic features of sounds by auditory neurons. We report here that auditory neurons can be classified into functional groups, each of which plays a specific role in extracting distinct complex sound features. We recorded the electrophysiological responses of single auditory neurons in the songbird midbrain and forebrain to conspecific song, measured their tuning by calculating spectrotemporal receptive fields (STRFs), and classified them using multiple cluster analysis methods. Based on STRF shape, cells clustered into functional groups that divided the space of acoustical features into regions that represent cues for the fundamental acoustic percepts of pitch, timbre, and rhythm. Four major groups were found in the midbrain, and five major groups were found in the forebrain. Comparing STRFs in midbrain and forebrain neurons suggested that both inheritance and emergence of tuning properties occur as information ascends the auditory processing stream.
Collapse
|
46
|
Increasing spectrotemporal sound density reveals an octave-based organization in cat primary auditory cortex. J Neurosci 2008; 28:8885-96. [PMID: 18768682 DOI: 10.1523/jneurosci.2693-08.2008] [Citation(s) in RCA: 40] [Impact Index Per Article: 2.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/21/2022] Open
Abstract
Auditory neurons are likely adapted to process complex stimuli, such as vocalizations, which contain spectrotemporal modulations. However, basic properties of auditory neurons are often derived from tone pips presented in isolation, which lack spectrotemporal modulations. In this context, it is unclear how to deduce the functional role of auditory neurons from their tone pip-derived tuning properties. In this study, spectrotemporal receptive fields (STRFs) were obtained from responses to multi-tone stimulus ensembles differing in their average spectrotemporal density (i.e., number of tone pips per second). STRFs for different stimulus densities were derived from multiple single-unit activity (MUA) and local field potentials (LFPs), simultaneously recorded in primary auditory cortex of cats. Consistent with earlier studies, we found that the spectral bandwidth was narrower for MUA compared with LFPs. Both neural firing rate and LFP amplitude were reduced when the density of the stimulus ensemble increased. Surprisingly, we found that increasing the spectrotemporal sound density revealed with increasing clarity an over-representation of response peaks at frequencies of approximately 3, 5, 10, and 20 kHz, in both MUA- and LFP-derived STRFs. Although the decrease in spectral bandwidth and neural activity with increasing stimulus density can likely be accounted for by forward suppression, the mechanisms underlying the over-representation of the octave-spaced response peaks are unclear. Plausibly, the over-representation may be a functional correlate of the periodic pattern of corticocortical connections observed along the tonotopic axis of cat auditory cortex.
Collapse
|
47
|
Nagel KI, Doupe AJ. Organizing principles of spectro-temporal encoding in the avian primary auditory area field L. Neuron 2008; 58:938-55. [PMID: 18579083 DOI: 10.1016/j.neuron.2008.04.028] [Citation(s) in RCA: 69] [Impact Index Per Article: 4.3] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/30/2007] [Revised: 02/11/2008] [Accepted: 04/18/2008] [Indexed: 11/30/2022]
Abstract
The organization of postthalamic auditory areas remains unclear in many respects. Using a stimulus based on properties of natural sounds, we mapped spectro-temporal receptive fields (STRFs) of neurons in the primary auditory area field L of unanesthetized zebra finches. Cells were sensitive to only a subset of possible acoustic features: nearly all neurons were narrowly tuned along the spectral dimension, the temporal dimension, or both; broadly tuned and strongly orientation-sensitive cells were rare. At high stimulus intensities, neurons were sensitive to differences in sound energy along their preferred dimension, while at lower intensities, neurons behaved more like simple detectors. Finally, we found a systematic relationship between neurons' STRFs, their electrophysiological properties, and their location in field L input or output layers. These data suggest that spectral and temporal processing are segregated within field L, and provide a unifying account of how field L response properties depend on stimulus intensity.
Collapse
Affiliation(s)
- Katherine I Nagel
- Keck Center for Integrative Neuroscience, Department of Physiology, University of California, San Francisco, San Francisco, CA 94143, USA.
| | | |
Collapse
|
48
|
Binaural interactions shape binaural response structures and frequency response functions in primary auditory cortex. Hear Res 2008; 238:68-76. [PMID: 18295994 DOI: 10.1016/j.heares.2008.01.003] [Citation(s) in RCA: 10] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 07/02/2007] [Revised: 01/09/2008] [Accepted: 01/15/2008] [Indexed: 11/24/2022]
Abstract
The overall purpose of this study is to examine the behavior of primary auditory cortex (AI) units in the three-dimensional stimulus space that resembles normal listening conditions, viz., level at the two ears and frequency. A binaural-level response area (LRA) is the response to a matrix of contralateral and ipsilateral stimuli presented at a single frequency. LRAs have been examined in the inferior colliculus and AI and found to be highly organized response patterns that are shaped by binaural interactions. The aggregate of LRAs across frequency is the binaural response structure (BRS), a new concept that captures unit behavior in this three-dimensional stimulus space. Since binaural interactions contribute greatly to configuring component LRAs, it is clear that binaural interactions help shape the aggregate BRS. The BRS contains the data required to generate binaural frequency response functions. The frequency range and magnitude of these functions depend on the level of the stimulus at each ear and the configuration of the BRS. Changing either level can greatly alter the binaural frequency response function. Thus, in addition to their classic role in localization, binaural interactions play a fundamentally important role in determining the frequency domain of units in AI.
Collapse
|
49
|
Christianson GB, Sahani M, Linden JF. The consequences of response nonlinearities for interpretation of spectrotemporal receptive fields. J Neurosci 2008; 28:446-55. [PMID: 18184787 PMCID: PMC6670552 DOI: 10.1523/jneurosci.1775-07.2007] [Citation(s) in RCA: 97] [Impact Index Per Article: 6.1] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/19/2007] [Revised: 10/07/2007] [Accepted: 11/08/2007] [Indexed: 11/21/2022] Open
Abstract
Neurons in the central auditory system are often described by the spectrotemporal receptive field (STRF), conventionally defined as the best linear fit between the spectrogram of a sound and the spike rate it evokes. An STRF is often assumed to provide an estimate of the receptive field of a neuron, i.e., the spectral and temporal range of stimuli that affect the response. However, when the true stimulus-response function is nonlinear, the STRF will be stimulus dependent, and changes in the stimulus properties can alter estimates of the sign and spectrotemporal extent of receptive field components. We demonstrate analytically and in simulations that, even when uncorrelated stimuli are used, interactions between simple neuronal nonlinearities and higher-order structure in the stimulus can produce STRFs that show contributions from time-frequency combinations to which the neuron is actually insensitive. Only when spectrotemporally independent stimuli are used does the STRF reliably indicate features of the underlying receptive field, and even then it provides only a conservative estimate. One consequence of these observations, illustrated using natural stimuli, is that a stimulus-induced change in an STRF could arise from a consistent but nonlinear neuronal response to stimulus ensembles with differing higher-order dependencies. Thus, although the responses of higher auditory neurons may well involve adaptation to the statistics of different stimulus ensembles, stimulus dependence of STRFs alone, or indeed of any overly constrained stimulus-response mapping, cannot demonstrate the nature or magnitude of such effects.
Collapse
Affiliation(s)
| | | | - Jennifer F. Linden
- UCL Ear Institute
- Department of Anatomy and Developmental Biology, University College London, London, WC1E 6BT, United Kingdom
| |
Collapse
|
50
|
Bandyopadhyay S, Reiss LAJ, Young ED. Receptive field for dorsal cochlear nucleus neurons at multiple sound levels. J Neurophysiol 2007; 98:3505-15. [PMID: 17898144 DOI: 10.1152/jn.00539.2007] [Citation(s) in RCA: 10] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/22/2022] Open
Abstract
Neurons in the dorsal cochlear nucleus (DCN) exhibit nonlinearities in spectral processing, which make it difficult to predict the neurons' responses to stimuli. Here, we consider two possible sources of nonlinearity: nonmonotonic responses as sound level increases due to inhibition and interactions between frequency components. A spectral weighting function model of rate responses is used; the model approximates the neuron's rate response as a weighted sum of the frequency components of the stimulus plus a second-order sum that captures interactions between frequencies. Such models approximate DCN neurons well at low spectral contrast, i.e., when the SD (contrast) of the stimulus spectrum is limited to 3 dB. This model is compared with a first-order sum with weights that are explicit functions of sound level, so that the low-contrast model is extended to spectral contrasts of 12 dB, the range of natural stimuli. The sound-level-dependent weights improve prediction performance at large spectral contrast. However, the interactions between frequencies, represented as second-order terms, are more important at low spectral contrast. The level-dependent model is shown to predict previously described patterns of responses to spectral edges, showing that small changes in the inhibitory components of the receptive field can produce large changes in the responses of the neuron to features of natural stimuli. These results provide an effective way of characterizing nonlinear auditory neurons incorporating stimulus-dependent sensitivity changes. Such models could be used for neurons in other sensory systems that show similar effects.
Collapse
Affiliation(s)
- Sharba Bandyopadhyay
- Center for Hearing and Balance and Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD 21205, USA
| | | | | |
Collapse
|