Impact of Feature Selection Algorithm on Speech Emotion Recognition Using Deep Convolutional Neural Network

Cross-corpus speech emotion recognition with transformers: Leveraging handcrafted features and data augmentation

Speech emotion recognition (SER) stands as a prominent and dynamic research field in data science due to its extensive application in various domains such as psychological assessment, mobile services, and computer games, mobile services. In previous research, numerous studies utilized manually engineered features for emotion classification, resulting in commendable accuracy. However, these features tend to underperform in complex scenarios, leading to reduced classification accuracy. These scenarios include: 1. Datasets that contain diverse speech patterns, dialects, accents, or variations in emotional expressions. 2. Data with background noise. 3. Scenarios where the distribution of emotions varies significantly across datasets can be challenging. 4. Combining datasets from different sources introduce complexities due to variations in recording conditions, data quality, and emotional expressions. Consequently, there is a need to improve the classification performance of SER techniques. To address this, a novel SER framework was introduced in this study. Prior to feature extraction, signal preprocessing and data augmentation methods were applied to augment the available data, resulting in the derivation of 18 informative features from each signal. The discriminative feature set was obtained using feature selection techniques which was then utilized as input for emotion recognition using the SAVEE, RAVDESS, and EMO-DB datasets. Furthermore, this research also implemented a cross-corpus model that incorporated all speech files related to common emotions from three datasets. The experimental outcomes demonstrated the superior performance of SER framework compared to existing frameworks in the field. Notably, the framework presented in this study achieved remarkable accuracy rates across various datasets. Specifically, the proposed model obtained an accuracy of 95%, 94%,97%, and 97% on SAVEE, RAVDESS, EMO-DB and cross-corpus datasets respectively. These results underscore the significant contribution of our proposed framework to the field of SER.

Human-Computer Interaction with a Real-Time Speech Emotion Recognition with Ensembling Techniques 1D Convolution Neural Network and Attention

Emotions have a crucial function in the mental existence of humans. They are vital for identifying a person's behaviour and mental condition. Speech Emotion Recognition (SER) is extracting a speaker's emotional state from their speech signal. SER is a growing discipline in human-computer interaction, and it has recently attracted more significant interest. This is because there are not so many universal emotions; therefore, any intelligent system with enough computational capacity can educate itself to recognise them. However, the issue is that human speech is immensely diverse, making it difficult to create a single, standardised recipe for detecting hidden emotions. This work attempted to solve this research difficulty by combining a multilingual emotional dataset with building a more generalised and effective model for recognising human emotions. A two-step process was used to develop the model. The first stage involved the extraction of features, and the second stage involved the classification of the features that were extracted. ZCR, RMSE, and the renowned MFC coefficients were retrieved as features. Two proposed models, 1D CNN combined with LSTM and attention and a proprietary 2D CNN architecture, were used for classification. The outcomes demonstrated that the suggested 1D CNN with LSTM and attention performed better than the 2D CNN. For the EMO-DB, SAVEE, ANAD, and BAVED datasets, the model's accuracy was 96.72%, 97.13%, 96.72%, and 88.39%, respectively. The model beat several earlier efforts on the same datasets, demonstrating the generality and efficacy of recognising multiple emotions from various languages.

Speech Emotion Recognition Based on Modified ReliefF

As the key of human-computer natural interaction, the research of emotion recognition is of great significance to the development of computer intelligence. In view of the issue that the current emotional feature dimension is too high, which affects the classification performance, this paper proposes a modified ReliefF feature selection algorithm to screen out feature subsets with smaller dimensions and better performance from high-dimensional features to further improve the efficiency and accuracy of emotion recognition. In the modified algorithm, the selection range of random samples is adjusted; the correlation between features is measured by the maximum information coefficient, and the distance measurement method between samples is established based on the correlation. The experimental results on the eNTERFACE'05 and SAVEE speech emotional datasets show that the features filtered based on the modified algorithm significantly reduce the data dimensions and effectively improve the accuracy of emotion recognition.

Recognition of Emotion with Intensity from Speech Signal Using 3D Transformed Feature and Deep Learning

Speech Emotion Recognition (SER), the extraction of emotional features with the appropriate classification from speech signals, has recently received attention for its emerging social applications. Emotional intensity (e.g., Normal, Strong) for a particular emotional expression (e.g., Sad, Angry) has a crucial influence on social activities. A person with intense sadness or anger may fall into severe disruptive action, eventually triggering a suicidal or devastating act. However, existing Deep Learning (DL)-based SER models only consider the categorization of emotion, ignoring the respective emotional intensity, despite its utmost importance. In this study, a novel scheme for Recognition of Emotion with Intensity from Speech (REIS) is developed using the DL model by integrating three speech signal transformation methods, namely Mel-frequency Cepstral Coefficient (MFCC), Short-time Fourier Transform (STFT), and Chroma STFT. The integrated 3D form of transformed features from three individual methods is fed into the DL model. Moreover, under the proposed REIS, both the single and cascaded frameworks with DL models are investigated. A DL model consists of a 3D Convolutional Neural Network (CNN), Time Distribution Flatten (TDF) layer, and Bidirectional Long Short-term Memory (Bi-LSTM) network. The 3D CNN block extracts convolved features from 3D transformed speech features. The convolved features were flattened through the TDF layer and fed into Bi-LSTM to classify emotion with intensity in a single DL framework. The 3D transformed feature is first classified into emotion categories in the cascaded DL framework using a DL model. Then, using a different DL model, the intensity level of the identified categories is determined. The proposed REIS has been evaluated on the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) benchmark dataset, and the cascaded DL framework is found to be better than the single DL framework. The proposed REIS method has shown remarkable recognition accuracy, outperforming related existing methods.

Real-Time End-to-End Speech Emotion Recognition with Cross-Domain Adaptation

Language resources are the main factor in speech-emotion-recognition (SER)-based deep learning models. Thai is a low-resource language that has a smaller data size than high-resource languages such as German. This paper describes the framework of using a pretrained-model-based front-end and back-end network to adapt feature spaces from the speech recognition domain to the speech emotion classification domain. It consists of two parts: a speech recognition front-end network and a speech emotion recognition back-end network. For speech recognition, Wav2Vec2 is the state-of-the-art for high-resource languages, while XLSR is used for low-resource languages. Wav2Vec2 and XLSR have proposed generalized end-to-end learning for speech understanding based on the speech recognition domain as feature space representations from feature encoding. This is one reason why our front-end network was selected as Wav2Vec2 and XLSR for the pretrained model. The pre-trained Wav2Vec2 and XLSR are used for front-end networks and fine-tuned for specific languages using the Common Voice 7.0 dataset. Then, feature vectors of the front-end network are input for back-end networks; this includes convolution time reduction (CTR) and linear mean encoding transformation (LMET). Experiments using two different datasets show that our proposed framework can outperform the baselines in terms of unweighted and weighted accuracies.

Automatic Speech Emotion Recognition of Younger School Age Children

This paper introduces the extended description of a database that contains emotional speech in the Russian language of younger school age (8–12-year-old) children and describes the results of validation of the database based on classical machine learning algorithms, such as Support Vector Machine (SVM) and Multi-Layer Perceptron (MLP). The validation is performed using standard procedures and scenarios of the validation similar to other well-known databases of children’s emotional acting speech. Performance evaluation of automatic multiclass recognition on four emotion classes “Neutral (Calm)—Joy—Sadness—Anger” shows the superiority of SVM performance and also MLP performance over the results of perceptual tests. Moreover, the results of automatic recognition on the test dataset which was used in the perceptual test are even better. These results prove that emotions in the database can be reliably recognized both by experts and automatically using classical machine learning algorithms such as SVM and MLP, which can be used as baselines for comparing emotion recognition systems based on more sophisticated modern machine learning methods and deep neural networks. The results also confirm that this database can be a valuable resource for researchers studying affective reactions in speech communication during child-computer interactions in the Russian language and can be used to develop various edutainment, health care, etc. applications.

Human-Computer Interaction with Detection of Speaker Emotions Using Convolution Neural Networks

Emotions play an essential role in human relationships, and many real-time applications rely on interpreting the speaker’s emotion from their words. Speech emotion recognition (SER) modules aid human-computer interface (HCI) applications, but they are challenging to implement because of the lack of balanced data for training and clarity about which features are sufficient for categorization. This research discusses the impact of the classification approach, identifying the most appropriate combination of features and data augmentation on speech emotion detection accuracy. Selection of the correct combination of handcrafted features with the classifier plays an integral part in reducing computation complexity. The suggested classification model, a 1D convolutional neural network (1D CNN), outperforms traditional machine learning approaches in classification. Unlike most earlier studies, which examined emotions primarily through a single language lens, our analysis looks at numerous language data sets. With the most discriminating features and data augmentation, our technique achieves 97.09%, 96.44%, and 83.33% accuracy for the BAVED, ANAD, and SAVEE data sets, respectively.

A DCRNN-based ensemble classifier for speech emotion recognition in Odia language

The Odia language is an old Eastern Indo-Aryan language, spoken by 46.8 million people across India. We have designed an ensemble classifier using Deep Convolutional Recurrent Neural Network for Speech Emotion Recognition (SER). This study presents a new approach for SER tasks motivated by recent research on speech emotion recognition. Initially, we extract utterance-level log Mel-spectrograms and their first and second derivative (Static, Delta, and Delta-delta), represented as 3-D log Mel-spectrograms. We utilize deep convolutional neural networks deep convolutional neural networks to extract the deep features from 3-D log Mel-spectrograms. Then a bi-directional-gated recurrent unit network is applied to express long-term temporal dependency out of all features to produce utterance-level emotion. Finally, we use ensemble classifiers using Softmax and Support Vector Machine classifier to improve the final recognition rate. In this way, our proposed framework is trained and tested on Odia (Seven emotional states) and RAVDESS (Eight emotional states) dataset. The experimental results reveal that an ensemble classifier performs better instead of a single classifier. The accuracy levels reached are 85.31% and 77.54%, outperforming some state-of-the-art frameworks on the Odia and RAVDESS datasets.

The Sustainable Development of Intangible Cultural Heritage with AI: Cantonese Opera Singing Genre Classification Based on CoGCNet Model in China

Chinese Cantonese opera, a UNESCO Intangible Cultural Heritage (ICH) of Humanity, has faced a series of development problems due to diversified entertainment and emerging cultures. While, the management on Cantonese opera data in a scientific manner is conducive to the sustainable development of ICH. Therefore, in this study, a scientific and standardized audio database dedicated to Cantonese opera is established, and a classification method for Cantonese opera singing genres based on the Cantonese opera Genre Classification Networks (CoGCNet) model is proposed given the similarity of the rhythm characteristics of different Cantonese opera singing genres. The original signal of Cantonese opera singing is pre-processed to obtain the Mel-Frequency Cepstrum as the input of the model. The cascade fusion CNN combines each segment’s shallow and deep features; the double-layer LSTM and CNN hybrid network enhance the contextual relevance between signals. This achieves intelligent classification management of Cantonese opera data, meanwhile effectively solving the problem that existing methods are difficult to classify accurately. Experimental results on the customized Cantonese opera dataset show that the method has high classification accuracy with 95.69% Precision, 95.58% Recall and 95.60% F1 value, and the overall performance is better than that of the commonly used neural network models. In addition, this method also provides a new feasible idea for the sustainable development of the study on the singing characteristics of the Cantonese opera genres.

Vector learning representation for generalized speech emotion recognition

•

A verify-to-classify framework was designed for achieving in generalization and overall performance.

•

An implemented verify-to-classify framework can work well in both verification (in-domain) and recognition (out-domain).

•

Our softmax with Lo5 can work well with emotion vectors and help improve classification performance.

Speech emotion recognition (SER) plays an important role in global business today to improve service efficiency. In the literature of SER, many techniques have been using deep learning to extract and learn features. Recently, we have proposed end-to-end learning for a deep residual local feature learning block (DeepResLFLB). The advantages of end-to-end learning are low engineering effort and less hyperparameter tuning. Nevertheless, this learning method is easily to fall into an overfitting problem. Therefore, this paper described the concept of the “verify-to-classify” framework to apply for learning vectors, extracted from feature spaces of emotional information. This framework consists of two important portions: speech emotion learning and recognition. In speech emotion learning, consisting of two steps: speech emotion verification enrolled training and prediction, the residual learning (ResNet) with squeeze-excitation (SE) block was used as a core component of both steps to extract emotional state vectors and build an emotion model by the speech emotion verification enrolled training. Then the in-domain pre-trained weights of the emotion trained model are transferred to the prediction step. As a result of the speech emotion learning, the accepted model—validated by EER—is transferred to the speech emotion recognition in terms of out-domain pre-trained weights, which are ready for classification using a classical ML method. In this manner, a suitable loss function is important to work with emotional vectors. Here, two loss functions were proposed: angular prototypical and softmax with angular prototypical losses. Based on two publicly available datasets: Emo-DB and RAVDESS, both with high- and low-quality environments. The experimental results show that our proposed method can significantly improve generalized performance and explainable emotion results, when evaluated by standard metrics: EER, accuracy, precision, recall, and F1-score.

Automated emotion recognition: Current trends and future perspectives

BACKGROUND

Human emotions greatly affect the actions of a person. The automated emotion recognition has applications in multiple domains such as health care, e-learning, surveillance, etc. The development of computer-aided diagnosis (CAD) tools has led to the automated recognition of human emotions.

OBJECTIVE

This review paper provides an insight into various methods employed using electroencephalogram (EEG), facial, and speech signals coupled with multi-modal emotion recognition techniques. In this work, we have reviewed most of the state-of-the-art papers published on this topic.

METHOD

This study was carried out by considering the various emotion recognition (ER) models proposed between 2016 and 2021. The papers were analysed based on methods employed, classifier used and performance obtained.

RESULTS

There is a significant rise in the application of deep learning techniques for ER. They have been widely applied for EEG, speech, facial expression, and multimodal features to develop an accurate ER model.

CONCLUSION

Our study reveals that most of the proposed machine and deep learning-based systems have yielded good performances for automated ER in a controlled environment. However, there is a need to obtain high performance for ER even in an uncontrolled environment.

Emotional Speech Recognition Using Deep Neural Networks

The expression of emotions in human communication plays a very important role in the information that needs to be conveyed to the partner. The forms of expression of human emotions are very rich. It could be body language, facial expressions, eye contact, laughter, and tone of voice. The languages of the world’s peoples are different, but even without understanding a language in communication, people can almost understand part of the message that the other partner wants to convey with emotional expressions as mentioned. Among the forms of human emotional expression, the expression of emotions through voice is perhaps the most studied. This article presents our research on speech emotion recognition using deep neural networks such as CNN, CRNN, and GRU. We used the Interactive Emotional Dyadic Motion Capture (IEMOCAP) corpus for the study with four emotions: anger, happiness, sadness, and neutrality. The feature parameters used for recognition include the Mel spectral coefficients and other parameters related to the spectrum and the intensity of the speech signal. The data augmentation was used by changing the voice and adding white noise. The results show that the GRU model gave the highest average recognition accuracy of 97.47%. This result is superior to existing studies on speech emotion recognition with the IEMOCAP corpus.

Establishment and psychometric characteristics of emotional words list for suicidal risk assessment in speech emotion recognition

BACKGROUND

Emotional disturbance is an important risk factor of suicidal behaviors. To ensure speech emotion recognition (SER), a novel technique to evaluate emotional characteristics of speech, precision in labeling emotional words is a prerequisite. Currently, a list of suicide-related emotional word is absent. The aims of this study were to establish an Emotional Words List for Suicidal Risk Assessment (EWLSRA) and test the reliability and validity of the list in a suicide-related SER task.

METHODS

Suicide-related emotion words were nominated and discussed by 10 suicide prevention professionals. Sixty-five tape-recordings of calls to a large psychological support hotline in China were selected to test psychometric characteristics of the EWLSRA.

RESULTS

The results shows that the EWLSRA consists of 11 emotion words which were highly associated with suicide risk scores and suicide attempts. Results of exploratory factor analysis support one-factor model of this list. The Fleiss' Kappa value of 0.42 indicated good inter-rater reliability of the list. In terms of criteria validities, indices of despair (Spearman ρ = 0.54, P < 0.001), sadness (ρ = 0.37, P = 0.006), helplessness (ρ = 0.45, P = 0.001), and numbness (ρ = 0.35, P = 0.009) were significantly associated with suicidal risk scores. The index of the emotional word of numbness in callers with suicide attempt during the 12-month follow-up was significantly higher than that in callers without suicide attempt during the follow-up (P = 0.049).

CONCLUSION

This study demonstrated that the EWLSRA has adequate psychometric performance in identifying suicide-related emotional words of recording of hotline callers to a national wide suicide prevention line. This list can be useful for SER in future studies on suicide prevention.

The Impact of Attention Mechanisms on Speech Emotion Recognition

Speech emotion recognition (SER) plays an important role in real-time applications of human-machine interaction. The Attention Mechanism is widely used to improve the performance of SER. However, the applicable rules of attention mechanism are not deeply discussed. This paper discussed the difference between Global-Attention and Self-Attention and explored their applicable rules to SER classification construction. The experimental results show that the Global-Attention can improve the accuracy of the sequential model, while the Self-Attention can improve the accuracy of the parallel model when conducting the model with the CNN and the LSTM. With this knowledge, a classifier (CNN-LSTM×2+Global-Attention model) for SER is proposed. The experiments result show that it could achieve an accuracy of 85.427% on the EMO-DB dataset.

Incorporating Interpersonal Synchronization Features for Automatic Emotion Recognition from Visual and Audio Data during Communication

During social interaction, humans recognize others’ emotions via individual features and interpersonal features. However, most previous automatic emotion recognition techniques only used individual features—they have not tested the importance of interpersonal features. In the present study, we asked whether interpersonal features, especially time-lagged synchronization features, are beneficial to the performance of automatic emotion recognition techniques. We explored this question in the main experiment (speaker-dependent emotion recognition) and supplementary experiment (speaker-independent emotion recognition) by building an individual framework and interpersonal framework in visual, audio, and cross-modality, respectively. Our main experiment results showed that the interpersonal framework outperformed the individual framework in every modality. Our supplementary experiment showed—even for unknown communication pairs—that the interpersonal framework led to a better performance. Therefore, we concluded that interpersonal features are useful to boost the performance of automatic emotion recognition tasks. We hope to raise attention to interpersonal features in this study.

Utterance Level Feature Aggregation with Deep Metric Learning for Speech Emotion Recognition

Emotion is a form of high-level paralinguistic information that is intrinsically conveyed by human speech. Automatic speech emotion recognition is an essential challenge for various applications; including mental disease diagnosis; audio surveillance; human behavior understanding; e-learning and human-machine/robot interaction. In this paper, we introduce a novel speech emotion recognition method, based on the Squeeze and Excitation ResNet (SE-ResNet) model and fed with spectrogram inputs. In order to overcome the limitations of the state-of-the-art techniques, which fail in providing a robust feature representation at the utterance level, the CNN architecture is extended with a trainable discriminative GhostVLAD clustering layer that aggregates the audio features into compact, single-utterance vector representation. In addition, an end-to-end neural embedding approach is introduced, based on an emotionally constrained triplet loss function. The loss function integrates the relations between the various emotional patterns and thus improves the latent space data representation. The proposed methodology achieves 83.35% and 64.92% global accuracy rates on the RAVDESS and CREMA-D publicly available datasets, respectively. When compared with the results provided by human observers, the gains in global accuracy scores are superior to 24%. Finally, the objective comparative evaluation with state-of-the-art techniques demonstrates accuracy gains of more than 3%.

Hybrid Method of Automated EEG Signals' Selection Using Reversed Correlation Algorithm for Improved Classification of Emotions

Based on the growing interest in encephalography to enhance human-computer interaction (HCI) and develop brain-computer interfaces (BCIs) for control and monitoring applications, efficient information retrieval from EEG sensors is of great importance. It is difficult due to noise from the internal and external artifacts and physiological interferences. The enhancement of the EEG-based emotion recognition processes can be achieved by selecting features that should be taken into account in further analysis. Therefore, the automatic feature selection of EEG signals is an important research area. We propose a multistep hybrid approach incorporating the Reversed Correlation Algorithm for automated frequency band-electrode combinations selection. Our method is simple to use and significantly reduces the number of sensors to only three channels. The proposed method has been verified by experiments performed on the DEAP dataset. The obtained effects have been evaluated regarding the accuracy of two emotions-valence and arousal. In comparison to other research studies, our method achieved classification results that were 4.20-8.44% greater. Moreover, it can be perceived as a universal EEG signal classification technique, as it belongs to unsupervised methods.