Speech emotion recognition(SER) often faces challenges due to the lack of large, annotated datasets. The presence of abundance unlabeled data offers a chance to explore methods that could significantly improve SER systems. This study explores the feasibility of enhancing general speech models by incorporating unimodal and multimodal training objectives derived from unlabeled data, specifically tailored to extract emotional content. These multimodal objectives aim to refineself-supervised learning(SSL)-based representations that, while effective in SER, were not originally created to extract emotional cues from speech. Our methodology introduces a set of multimodal objectives focused on capturing information from three primary sources: acoustic signals, through a representation objective based on the extendedGeneva Minimalistic Acoustic Parameter Set(eGEMAPS); facial expressions, via visual representations obtained from a pre-trained facial expression recognition system; and textual content, through pseudo-labels generated by a pre-trained emotion sentiment model. These objectives are automatically generated from 70.7 hours of unlabeled emotional content captured in naturalistic settings. We apply our strategy to four state-of-the-art SSL-based speech models, aiming to enhance their capabilities in SER tasks with multimodal signals while still keeping inference strictly audio-only. Our experimental evaluations across the CREMA-D, MSP-IMPROV, and MSP-Podcast datasets demonstrate that our approach significantly improves SER performance, especially in settings with limited labeled data.

Emotion recognition is inherently a multimodal problem. Humans use both audible and visual cues to determine a person’s emotions. There has been extensive improvement in the methods we use to fuse audio and visual representations between two unimodal deep-learning models. However, there is a lack of accommodation for modalities that have a disparity in the amount of computational resources needed to provide the same amount of temporal information. As the sequence length increases, current methods often make simplifications such as discarding frames or cropping the sequence. This paper introduces a chunking methodology designed for cross-attention-based multimodal transformer architectures. The approach involves segmenting the visual input—the more computationally demanding modality—into chunks. Cross-attention is then performed between the encoded audio and visual features instead of the original sequence lengths of the unimodal backbones. Our method achieves significant improvements over conventional cross-attention techniques in the audio-visual domain for a six-class emotional recognition problem, demonstrating better F1 score, precision, and recall on the CREMA-D database while reducing computational overhead.

Audio-Visual Speech-to-Speech Translation (AVS2S) typically prioritizes improving translation quality and naturalness. However, an equally critical aspect in audio-visual content is lip-synchrony—ensuring that the movements of the lips match the spoken content—essential for maintaining realism in dubbed videos. Despite its importance, the inclusion of lip-synchrony constraints in AVS2S models has been largely overlooked. This study addresses this gap by integrating a lip-synchrony loss into the training process of AVS2S models. Our proposed method significantly enhances lip-synchrony in direct audio-visual speechto-speech translation, achieving an average LSE-D score of 10.67, representing a 9.2% reduction in LSE-D over a strong baseline across four language pairs. Additionally, it maintains the naturalness and high quality of the translated speech when overlaid onto the original video, without any degradation in translation quality.

When selecting test data for subjective tasks, most studies define ground truth labels using aggregation methods such as the majority or plurality rules. These methods discard data points without consensus, making the test set easier than practical tasks where a prediction is needed for each sample. However, the discarded data points often express ambiguous cues that elicit coexisting traits perceived by annotators. This paper addresses the importance of considering all the annotations and samples in the data, highlighting that only showing the model's performance on an incomplete test set selected by using the majority or plurality rules can lead to bias in the models’ performances. We focus onspeech-emotion recognition(SER) tasks. We observe that traditional aggregation rules have a data loss ratio ranging from 5.63% to 89.17%. From this observation, we propose a flexible method named the all-inclusive aggregation rule to evaluate SER systems on the complete test data. We contrast traditional single-label formulations with a multi-label formulation to consider the coexistence of emotions. We show that training an SER model with the data selected by the all-inclusive aggregation rule shows consistently higher macro-F1 scores when tested in the entire test set, including ambiguous samples without agreement.

Most current audio-visual emotion recognition models lack the flexibility needed for deployment in practical applications. We envision a multimodal system that works even when only one modality is available and can be implemented interchangeably for either predicting emotional attributes or recognizing categorical emotions. Achieving such flexibility in a multimodal emotion recognition system is difficult due to the inherent challenges in accurately interpreting and integrating varied data sources. It is also a challenge to robustly handle missing or partial information while allowing direct switch between regression or classification tasks. This study proposes a versatile audio-visual learning (VAVL) framework for handling unimodal and multimodal systems for emotion regression or emotion classification tasks. We implement an audio-visual framework that can be trained even when audio and visual paired data is not available for part of the training set (i.e., audio only or only video is present). We achieve this effective representation learning with audio-visual shared layers, residual connections over shared layers, and a unimodal reconstruction task. Our experimental results reveal that our architecture significantly outperforms strong baselines on the CREMA-D, MSP-IMPROV, and CMU-MOSEI corpora. Notably, VAVL attains a new state-of-the-art performance in the emotional attribute prediction task on the MSP-IMPROV corpus.

Speech Emotion Recognition (SER) faces a distinct challenge compared to other speech-related tasks because the annotations will show the subjective emotional perceptions of different annotators. Previous SER studies often view the subjectivity of emotion perception as noise by using the majority rule or plurality rule to obtain the consensus labels. However, these standard approaches overlook the valuable information of labels that do not agree with the consensus and make it easier for the test set. Emotion perception can have co-occurring emotions in realistic conditions, and it is unnecessary to regard the disagreement between raters as noise. To bridge the SER into a multi-label task, we introduced an “all-inclusive rule,” which considers all available data, ratings, and distributional labels as multi-label targets and a complete test set. We demonstrated that models trained with multi-label targets generated by the proposed AR outperform conventional single-label methods across incomplete and complete test sets.

Speech emotion recognition (SER) is an essential technology for human-computer interaction systems. However, the previous study reveals that 80.77% of SER papers yield results that cannot be reproduced on the well-known IEMOCAP dataset. The main reason for reproducibility challenges is that the database did not provide standard data splits (e.g., train, development, and test sets). Prior papers could define its partition, but they did not provide details of the partition or source code for processing the partition. Therefore, this work aims to make SER open and reproducible to everyone. We develop the EMO-SUPERB, shorted for EMOtion Speech Universal PERformance Benchmark, including a user-friendly codebase to leverage 16 state-of-the-art (SOTA) speech self-supervised learning models for exhaustive evaluation plus one SOTA SER model across 6 open-source SER datasets in English and Chinese. We make all resources open-source to facilitate future developments in SER. Researchers can easily upload their systems or datasets to EMO-SUPERB, and we name the project “Open-Emotion”.

The field of speech emotion recognition (SER) aims to create scientifically rigorous systems that can reliably characterize emotional behaviors expressed in speech. A key aspect for building SER systems is to obtain emotional data that is both reliable and reproducible for practitioners. However, academic researchers encounter difficulties in accessing or collecting naturalistic large-scale, reliable emotional recordings. Also, the best practices for data collection are not necessarily described or shared when presenting emotional corpora. To address this issue, the paper proposes the creation of an affective naturalistic database consortium (AndC) that can encourage multidisciplinary cooperation among researchers and practitioners in the field of affective computing. This paper’s contribution is twofold. First, it proposes the design of the AndC with a customizable-standard framework for intelligently-controlled emotional data collection. The focus is on leveraging naturalistic spontaneous recordings available on audio-sharing websites. Second, it presents as a case study the development of a naturalistic large-scale Taiwanese Mandarin podcast corpus using the customizable-standard intelligently-controlled framework. The AndC will enable research groups to effectively collect data using the provided pipeline and to contribute with alternative algorithms or data collection protocols.

Representation learning is a challenging, but essential task in audiovisual learning. A key challenge is to generate strong cross-modal representations while still capturing discriminative information contained in unimodal features. Properly capturing this information is important to increase accuracy and robustness in audiovisual tasks. Focusing on emotion recognition, this study proposes novel cross-modal ladder networks to capture modality-specific information while building strong cross-modal representations. Our method utilizes representations from a backbone network to implement unsupervised auxiliary tasks to reconstruct intermediate layer representations across the acoustic and visual networks. The skip connections between the cross-modal encoder and decoder provide powerful modality-specific and multimodal representations for emotion recognition. Our model on the CREMA-D corpus achieves high performance with precision, recall, and F1 scores over 80% on a six-class problem.

A challenging task in audiovisual emotion recognition is to implement neural network architectures that can leverage and fuse multimodal information while temporally aligning modalities, handling missing modalities, and capturing information from all modalities without losing information during training. These requirements are important to achieve model robustness and to increase accuracy on the emotion recognition task. A recent approach to perform multimodal fusion is to use the transformer architecture to properly fuse and align the modalities. This study proposes the AuxFormer framework, which addresses in a principled way the aforementioned challenges. AuxFormer combines the transformer framework with auxiliary networks. It uses shared losses to infuse information from single-modality networks that are separately embedded. The extra layer of audiovisual information added to our main network retains information that would otherwise be lost during training. The results show that the AuxFormer architecture achieves macro and micro F1Scores of 71.3% and 71.7%, respectively, on the CREMA-D corpus. For the MSP-IMPROV corpus, AuxFormer achieves a macro and micro F1-Scores of 70.4% and 76.5%, respectively. The results for both corpora are significantly better than strong baselines, indicating that our framework benefits from auxiliary networks. We also show that under non-ideal conditions (e.g., missing modalities) our architecture is able to sustain strong performance under audio-only and video-only scenarios, benefiting from a optimized training strategy.

Emotion recognition using audiovisual features is a challenging task for human-machine interaction systems. Under ideal conditions (perfect illumination, clean speech signals, and non-occluded visual data) many systems are able to achieve reliable results. However, few studies have considered developing multimodal systems and training strategies to build systems that can perform well under non ideal conditions. Audiovisual models still face challenging problems such as misalignment of modalities, lack of temporal modeling, and missing features due to noise or occlusions. In this article, we implement a model that combines auxiliary networks, a transformer architecture, and an optimized training mechanism to achieve a robust system for audiovisual emotion recognition that addresses, in a principled way, these challenges. Our evaluation analyzes how well this model performs in ideal conditions and when modalities are missing. We contrast this method with other multimodal fusion methods for emotion recognition. Our experimental results based on two audiovisual databases demonstrate that the proposed framework achieves: 1) improvements in emotion recognition accuracy, 2) better alignment and fusion of audiovisual features at the model level, 3) awareness of temporal information, and 4) robustness to non-ideal scenarios.