Semi-Supervised Speech Emotion Recognition With Ladder Networks

In the field of speech emotion recognition(SER), great progress has been made in feature extraction and classifier construction. However, in the natural environment, speech emotion recognition is vulnerable to environmental noise. Moreover, cross-corpus SER is a thorny problem because the gender differences and language differences between corpus. Therefore, these problems need to be solved. Considering the noise problem, wavelet threshold denoising was proposed. After that, this paper used multi-task learning method to solve the problem of languages and gender differences. Emotion recognition is primary task, while gender recognition and language recognition are auxiliary tasks. According to the multi-task learning model, the shared LSTM is designed to extract the shared features. Three private LSTMs are used to extract the features of emotion recognition ,language recognition and gender recognition respectively. Orthogonal constraint training was used to make the private space contain only private features. Two experiments are designed in this paper. The first experiment is the comparison between different wavelet bases and different numbers of decomposition layers to select suitable wavelet base and the number of decomposition layer. The second experiment is the comparison between cross-corpus SER in other literature.

Although recent research on speech emotion recognition has demonstrated that learning domain-invariant features provide an elegant solution to domain mismatch, the features learned by the existing methods lack generalization capabilities to capture latent information from datasets. We propose two novel domain adaptation methods, the generalized domain adversarial neural network (GDANN) and the class-aligned GDANN (CGDANN), to learn generalized domain-invariant representations for emotion recognition. GDANN and CGDANN, which are derived from multitask learning (MTL), consist of three tasks. The main task is to recognize the emotional category to which the input belongs. The remaining two tasks are auxiliary tasks. One is to use a variational autoencoder to model the input distribution, which encourages the model to learn the distribution of latent representations. The other is to learn the common representations of different domains, for which distinguishing via the domain classifier is difficult. The gradient of the domain classifier guides the shared representations of the source and target domains to approximate each other using a gradient reversal layer. To evaluate the effectiveness of the proposed methods, we conduct several experiments with the IEMOCAP and MSP-IMPROV datasets. The results illustrate that good performance is achieved compared with that of state-of-the-art methods. Notably, CGDANN utilizes a small quantity of labeled target domain samples to align the distribution representation and obtains the best performance among the comparison methods. We further visualize the representations learned by the proposed methods and discover that the representations of the source and target domains converge with a low variance.

Multi-task learning (MTL) has achieved success over a wide range of problems, where the goal is to improve the performance of a primary task using a set of relevant auxiliary tasks. However, when the usefulness of the auxiliary tasks w.r.t. the primary task is not known a priori, the success of MTL models depends on the correct choice of these auxiliary tasks and also a balanced mixing ratio of these tasks during alternate training. These two problems could be resolved via manual intuition or hyper-parameter tuning over all combinatorial task choices, but this introduces inductive bias or is not scalable when the number of candidate auxiliary tasks is very large. To address these issues, we present AutoSeM, a two-stage MTL pipeline, where the first stage automatically selects the most useful auxiliary tasks via a Beta-Bernoulli multi-armed bandit with Thompson Sampling, and the second stage learns the training mixing ratio of these selected auxiliary tasks via a Gaussian Process based Bayesian optimization framework. We conduct several MTL experiments on the GLUE language understanding tasks, and show that our AutoSeM framework can successfully find relevant auxiliary tasks and automatically learn their mixing ratio, achieving significant performance boosts on several primary tasks. Finally, we present ablations for each stage of AutoSeM and analyze the learned auxiliary task choices.

Artificial intelligence (AI) and machine learning (ML) are employed to make systems smarter. Today, the speech emotion recognition (SER) system evaluates the emotional state of the speaker by investigating his/her speech signal. Emotion recognition is a challenging task for a machine. In addition, making it smarter so that the emotions are efficiently recognized by AI is equally challenging. The speech signal is quite hard to examine using signal processing methods because it consists of different frequencies and features that vary according to emotions, such as anger, fear, sadness, happiness, boredom, disgust, and surprise. Even though different algorithms are being developed for the SER, the success rates are very low according to the languages, the emotions, and the databases. In this paper, we propose a new lightweight effective SER model that has a low computational complexity and a high recognition accuracy. The suggested method uses the convolutional neural network (CNN) approach to learn the deep frequency features by using a plain rectangular filter with a modified pooling strategy that have more discriminative power for the SER. The proposed CNN model was trained on the extracted frequency features from the speech data and was then tested to predict the emotions. The proposed SER model was evaluated over two benchmarks, which included the interactive emotional dyadic motion capture (IEMOCAP) and the berlin emotional speech database (EMO-DB) speech datasets, and it obtained 77.01% and 92.02% recognition results. The experimental results demonstrated that the proposed CNN-based SER system can achieve a better recognition performance than the state-of-the-art SER systems.

Conventional models for emotion recognition from speech signal are trained in supervised fashion using speech utterances with emotion labels. In this study we hypothesize that speech signal depends on multiple latent variables including the emotional state, age, gender, and speech content. We propose an Adversarial Autoencoder (AAE) to perform variational inference over the latent variables and reconstruct the input feature representations. Reconstruction of feature representations is used as an auxiliary task to aid the primary emotion recognition task. Experiments on the IEMOCAP dataset demonstrate that the auxiliary learning tasks improve emotion classification accuracy compared to a baseline supervised classifier. Further, we demonstrate that the proposed learning approach can be used for the end-to-end speech emotion recognition, as its applicable for models that operate on frame-level inputs.

Multitask learning for acoustic scene classification with topic-based soft labels and a mutual attention mechanism

Context-aware computing has become an essential component of modern wearable technology, enabling devices to sense, interpret, and respond to users’ environments and affective states. By integrating contextual intelligence, smart systems can adapt their own behavior to better align with user needs, enhancing efficiency, personalization, and user experience. A critical aspect of context awareness is the ability to recognize a user’s internal state, particularly emotions, which influence decision-making, communication, and overall well-being. While emotion recognition is traditionally approached through multimodal sensing—using cameras, microphones, or physiological sensors such as Electroencephalography (EEG), Electrocardiography (ECG), and Electro-dermal Activity (EDA)—these methods often require multiple sensors, leading to challenges related to power consumption, complexity, and privacy. The demand for a more efficient and unobtrusive approach to emotion and context recognition has driven research toward minimal-sensor solutions that can extract meaningful affective and contextual information with reduced hardware requirements. Restricting sensing sources of contextual information by using only limited number of sensors might come at the expense of constraining the ultimate goal of context-awareness. Such expense can be mitigated by “doing more with less” i.e. delving into the opportunities of using a single sensor for multiple purposes. What is desirable is a minimal-sensor solution capable of capturing both motion and physiological signals for both robust emotion recognition and additional context estimation. Among wearable sensors, chest-worn Inertial Measurement Units (IMUs) stand out as promising candidates for comprehensive context estimation, as they are well-positioned to capture motion patterns, and physiological vibrations, being the subtle cardio-respiratory signals. This raises the question of whether a single chest-worn IMU could serve as a minimal yet effective sensor for simultaneously estimating both emotional and contextual information, offering a practical alternative to traditional multi-sensor approaches. To address the challenge of using chest-worn IMU for multiple context sensing, I investigated the potential of using Seismocardiography (SCG) and Accelerometry-Derived Respiration (ADR) to estimate emotional states. First, I explored the existing capabilities of chest-worn IMUs in capturing cardio-respiratory vibrations and motion data, which I used as a knowledge base for embedding such IMUs in context sensing. Data plays a crucial role in this problem, therefore I developed a flexible software interface, ColEmo, to facilitate the scalable collection of emotion-related data and introduced a novel dataset, EmoWear, integrating physiological and motion data for emotion recognition and context awareness. Then, I examined the feasibility of using SCG for emotion recognition, comparing its performance with traditional physiological signals and evaluating its effectiveness as a single-sensor solution. This thesis presents a detailed analysis of the potential of chest-worn IMUs for emotion and context recognition. It reports on the design and validation of a SCG-based wearable system, the development and characteristics of the EmoWear dataset, and the implementation of the ColEmo platform for emotion data collection. The research also examines machine learning techniques applied to the SCG and ADR signals, assessing their effectiveness in estimating emotional states. My findings indicate that SCG-based emotion recognition achieves performance comparable to conventional physiological counterparts, demonstrating the feasibility of using a single IMU sensor for affective computing. The publicly-available EmoWear dataset provides a valuable resource for future research, highlighting the potential of combining motion and physiological data for enhanced context awareness. Additionally, the ColEmo software interface, which I published open-source, proves to be an effective tool for multi-modal emotion data collection, facilitating further advancements in wearable emotion recognition. The results pave the way for a minimal-sensor approach for context information estimation that can significantly reduce system complexity while maintaining high performance. These findings confirm that chest-worn IMUs present an opportunity to further improve emotion recognition in addition to the motion-based context information estimation. My research highlights the need for standardized methodologies in affective computing and demonstrates the viability of motion-based physiological sensing for emotion estimation. By using a single chest-worn IMU, this work presents an alternative to multi-sensor systems, reducing power consumption, and privacy concerns while maintaining robust performance. Future research should focus on refining IMU-based emotion recognition using deep learning techniques and larger datasets to enhance classification accuracy. Further validation in real-world settings is necessary to assess the practical applicability of the proposed system. The EmoWear dataset should be explored for additional context-awareness applications, such as activity recognition and stress monitoring, to maximize its utility. The ColEmo platform offers opportunities for expansion beyond emotion recognition, including cognitive load estimation, stress monitoring, and human-computer interaction. This research moves wearable technology forward by using fewer sensors, helping make affective computing more efficient, versatile, and respectful of privacy.

Speech emotion recognition in real static and dynamic human-robot interaction scenarios

The acoustic cues that convey emotion in speech are similar to those that convey emotion in music, and recognition of emotion in both of these types of cue recruits overlapping networks in the brain. Given the similarities between music and speech prosody, developmental research is uniquely positioned to determine whether recognition of these cues develops in parallel. In the present study, we asked 60 children aged 6 to 11 years, and 51 university students, to judge the emotions of 10 musical excerpts, 10 inflected speech clips, and 10 affect burst clips. We presented stimuli intended to convey happiness, sadness, anger, fear, and pride. Each emotion was presented twice per type of stimulus. We found that recognition of emotions in music and speech developed in parallel, and adult-levels of recognition develop later for these stimuli than for affect bursts. We also found that sad stimuli were most easily recognised, followed by happiness, fear, and then anger. In addition, we found that recognition of emotion in speech and affect bursts can predict emotion recognition in music stimuli independently of age and musical training. Finally, although proud speech and affect bursts were not well recognised, children aged eight years and older showed adult-like responses in recognition of proud music.

To solve the problem of feature distribution discrepancy in cross-corpus speech emotion recognition tasks, this paper proposed an emotion recognition model based on multi-task learning and subdomain adaptation, which alleviates the impact on emotion recognition. Existing methods have shortcomings in speech feature representation and cross-corpus feature distribution alignment. The proposed model uses a deep denoising auto-encoder as a shared feature extraction network for multi-task learning, and the fully connected layer and softmax layer are added before each recognition task as task-specific layers. Subsequently, the subdomain adaptation algorithm of emotion and gender features is added to the shared network to obtain the shared emotion features and gender features of the source domain and target domain, respectively. Multi-task learning effectively enhances the representation ability of features, a subdomain adaptive algorithm promotes the migrating ability of features and effectively alleviates the impact of feature distribution differences in emotional features. The average results of six cross-corpus speech emotion recognition experiments show that, compared with other models, the weighted average recall rate is increased by 1.89~10.07%, the experimental results verify the validity of the proposed model.

Unsupervised feature selection and NMF de-noising for robust Speech Emotion Recognition

For speech emotion recognition, emotional feature set with high dimension may produce redundant features and influence the recognition accuracy. To solve this problem and obtain the optimal emotional feature subset of speech, a feature dimension reduction based on linear discriminant analysis is proposed. According to the confusion degree between different basic emotions, an emotion recognition method based on support vector machine decision tree is proposed. Experiment on speaker-dependent speech emotion recognition using Chinese speech database from institute of automation of Chinese academy of sciences is performed and a speech emotion recognition system is presented, where standard feature sets of the INTER-SPEECH and classic classifiers are used in comparative experiments respectively. Experimental results show that the proposal achieves 84.39% recognition accuracy on average. By proposal, it would be fast and efficient to discriminate emotional states of diverse speakers from speech, and it would make it possible to realize the interaction between speaker and computer/robot in the future.

The accurate recognition of emotions in conversations helps understand the speaker’s intentions and facilitates various analyses in artificial intelligence, especially in human–computer interaction systems. However, most previous methods need more ability to track the different emotional states of each speaker in a dialogue. To alleviate this dilemma, we propose a new approach, Multi-Task Learning and Multi-Fusion AudioText Emotion Recognition in Conversation (MMATERIC) for emotion recognition in conversation. MMATERIC can refer to and combine the benefits of two distinct tasks: emotion recognition in text and emotion recognition in speech, and production of fused multimodal features to recognize the emotions of different speakers in dialogue. At the core of MATTERIC are three modules: an encoder with multimodal attention, a speaker emotion detection unit (SED-Unit), and a decoder with speaker emotion detection Bi-LSTM (SED-Bi-LSTM). Together, these three modules model the changing emotions of a speaker at a given moment in a conversation. Meanwhile, we adopt multiple fusion strategies in different stages, mainly using model fusion and decision stage fusion to improve the model’s accuracy. Simultaneously, our multimodal framework allows features to interact across modalities and allows potential adaptation flows from one modality to another. Our experimental results on two benchmark datasets show that our proposed method is effective and outperforms the state-of-the-art baseline methods. The performance improvement of our method is mainly attributed to the combination of three core modules of MATTERIC and the different fusion methods we adopt in each stage.

Speech Emotion Recognition (SER) plays a vital role in various real-world applications, from mental health diagnostics to human-computer interaction. This research examines the integration of Stacked Long Short-Term Memory (LSTM) networks and Hidden Markov Models (HMM) for SER, addressing the limitations of traditional models like Gaussian Mixture Models (GMM) and Support Vector Machines (SVM). While GMMs and SVMs offer decent performance, their inability to model temporal dynamics limits their accuracy in detecting complex emotions. The proposed hybrid model leverages LSTM’s strength in handling sequential data and HMM’s ability to model emotional transitions, making it highly suitable for real-world noisy environments. Preprocessing techniques such as MFCC and LPCC are applied to enhance feature extraction, and benchmark datasets like IEMOCAP and RAVDESS are used for evaluation. Finally, The Systematic review highlights the superior role of the hybrid model's performance on SER and sets the stage for a significant shift in future research in addressing bias and fairness in SER systems by combining LSTM and HMM as hybrid model. In recent studies, context-aware emotion recognition models are commonly evaluated using existing datasets such as IEMOCAP, MELD, DailyDialog, and SEMAINE, which provide conversational and multimodal emotional data. Features used often include acoustic features as MFCCs, spectrograms, prosodic features, linguistic features as BERT or transformer embeddings, and visual cues such as facial landmarks or video frames. Many models also incorporate contextual embeddings from transformers, graph attention networks, or sequential memory modules to capture speaker history and emotional shifts.

Purpose of research.Emotions play one of the key roles in the regulation of human behaviour. Solving the problem of automatic recognition of emotions makes it possible to increase the effectiveness of operation of a whole range of digital systems such as security systems, human-machine interfaces, e-commerce systems, etc. At the same time, the low efficiency of modern approaches to recognizing emotions in speech can be noted. This work studies automatic recognition of emotions in speech applying machine learning methods.Methods.The article describes and tests an approach to automatic recognition of emotions in speech based on multitask learning of deep convolution neural networks of AlexNet and VGG architectures using automatic selection of the weight coefficients for each task when calculating the final loss value during learning. All the models were trained on a sample of the IEMOCAP dataset with four emotional categories of ‘anger’, ‘happiness’, ‘neutral emotion’, ‘sadness’. The log-mel spectrograms of statements processed by a specialized algorithm are used as input data.Results.The considered models were tested on the basis of numerical metrics: the share of correctly recognized instances, accuracy, completeness, f-measure. For all of the above metrics, an improvement in the quality of emotion recognition by the proposed model was obtained in comparison with the two basic single-task models as well as with known solutions. This result is achieved through the use of automatic weighting of the values of the loss functions from individual tasks when forming the final value of the error in the learning process.Conclusion.The resulting improvement in the quality of emotion recognition in comparison with the known solutions confirms the feasibility of applying multitask learning to increase the accuracy of emotion recognition models. The developed approach makes it possible to achieve a uniform and simultaneous reduction of errors of individual tasks, and is used in the field of emotions recognition in speech for the first time.