Predicting the quality of enhanced wideband speech with a cochlear model

An overview of the HASPI and HASQI metrics for predicting speech intelligibility and speech quality for normal hearing, hearing loss, and hearing aids

Alterations of the speech signal, including additive noise and nonlinear distortion, can reduce speech intelligibility and quality. Hearing aids present an especially complicated situation since these devices may implement nonlinear processing designed to compensate for the hearing loss. Hearing-aid processing is often realized as time-varying multichannel gain adjustments, and may also include frequency reassignment. The challenge in designing metrics for hearing aids and hearing-impaired listeners is to accurately model the perceptual trade-offs between speech audibility and the nonlinear distortion introduced by hearing-aid processing. This paper focuses on the Hearing Aid Speech Perception Index (HASPI) and the Hearing Aid Speech Quality Index (HASQI) as representative metrics for predicting intelligibility and quality. These indices start with a model of the auditory periphery that can be adjusted to represent hearing loss. The peripheral model, the speech features computed from the model outputs, and the procedures used to fit the features to subject data are described. Examples are then presented for using the metrics to measure the effects of additive noise, evaluate noise-suppression processing, and to measure the differences among commercial hearing aids. Open questions and considerations in using these and related metrics are then discussed.

Phenomenological model of auditory nerve population responses to cochlear implant stimulation

BACKGROUND

Models of auditory nerve fiber (ANF) responses to electrical stimulation are helpful to develop advanced coding for cochlear implants (CIs). A phenomenological model of ANF population responses to CI electrical stimulation with a lower computational complexity compared to a biophysical model would be beneficial to evaluate new CI coding strategies.

NEW METHOD

This study presents a phenomenological model which combines four temporal characteristics of ANFs (refractoriness, facilitation, accommodation and spike rate adaptation) in addition to a spatial spread of the electric field.

RESULTS

The model predicts the performances of CI subjects in the melodic contour identification (MCI) experiment. The simulations for the MCI experiment were consistent with CI recipients' experimental outcomes that were not predictable from the electrical stimulation patterns themselves.

COMPARISON WITH EXISTING METHODS

Previously, no phenomenological population model of ANFs has combined all four aforementioned temporal phenomena.

CONCLUSIONS

The proposed model would help the further investigations of ANFs responses to different electrical stimulation patterns and comparison of different sound coding strategies in CIs.

Simple transformations capture auditory input to cortex

Sounds are processed by the ear and central auditory pathway. These processing steps are biologically complex, and many aspects of the transformation from sound waveforms to cortical response remain unclear. To understand this transformation, we combined models of the auditory periphery with various encoding models to predict auditory cortical responses to natural sounds. The cochlear models ranged from detailed biophysical simulations of the cochlea and auditory nerve to simple spectrogram-like approximations of the information processing in these structures. For three different stimulus sets, we tested the capacity of these models to predict the time course of single-unit neural responses recorded in ferret primary auditory cortex. We found that simple models based on a log-spaced spectrogram with approximately logarithmic compression perform similarly to the best-performing biophysically detailed models of the auditory periphery, and more consistently well over diverse natural and synthetic sounds. Furthermore, we demonstrated that including approximations of the three categories of auditory nerve fiber in these simple models can substantially improve prediction, particularly when combined with a network encoding model. Our findings imply that the properties of the auditory periphery and central pathway may together result in a simpler than expected functional transformation from ear to cortex. Thus, much of the detailed biological complexity seen in the auditory periphery does not appear to be important for understanding the cortical representation of sound.

The dynamic gammawarp auditory filterbank

Auditory filterbanks are an integral part of many metrics designed to predict speech intelligibility and speech quality. Considerations in these applications include accurate reproduction of auditory filter shapes, the ability to reproduce the impact of hearing loss as well as normal hearing, and computational efficiency. This paper presents an alternative method for implementing a dynamic compressive gammachirp (dcGC) auditory filterbank [Irino and Patterson (2006). IEEE Trans. Audio Speech Lang. Proc. 14, 2222-2232]. Instead of using a cascade of second-order sections, this approach uses digital frequency warping to give the gammawarp filterbank. The set of warped finite impulse response filter coefficients is constrained to be symmetrical, which results in the same phase response for all filters in the filterbank. The identical phase responses allow the dynamic variation in the gammachirp filter magnitude response to be realized as a sum, using time-varying weights, of three filters that provide the responses for high-, mid-, and low-intensity input signals, respectively. The gammawarp filterbank offers a substantial improvement in execution speed compared to previous dcGC implementations; for a dcGC filterbank, the gammawarp implementation is 24 to 38 times faster than the dcGC Matlab code of Irino.