Explainable AI (XAI) for Speech Emotion Recognition (SER) is critical for building transparent, trustworthy models. Current saliency-based methods, adapted from vision, highlight spectrogram regions but fail to show whether these regions correspond to meaningful acoustic markers of emotion, limiting faithfulness and interpretability. We propose a framework that overcomes these limitations by quantifying the magnitudes of cues within salient regions. This clarifies "what" is highlighted and connects it to "why" it matters, linking saliency to expert-referenced acoustic cues of speech emotions. Experiments on benchmark SER datasets show that our approach improves explanation quality by explicitly linking salient regions to theory-driven speech emotions expert-referenced acoustics. Compared to standard saliency methods, it provides more understandable and plausible explanations of SER models, offering a foundational step towards trustworthy speech-based affective computing.

Abstract--Understanding the nuances of speech emotion dataset curation and labeling is essential for assessing speech emotion recognition (SER) model potential in real-world applications. Most training and evaluation datasets contain acted or pseudo-acted speech (e.g., podcast speech) in which emotion expressions may be exaggerated or otherwise intentionally modified. Furthermore, datasets labeled based on crowd perception often lack transparency regarding the guidelines given to annotators. These factors make it difficult to understand model performance and pinpoint necessary areas for improvement. T o address this gap, we identified the Switchboard corpus as a promising source of naturalistic conversational speech, and we trained a crowd to label the dataset for categorical emotions (anger, contempt, disgust, fear, sadness, surprise, happiness, tenderness, calmness, and neutral) and dimensional attributes (activation, valence, and dominance). We refer to this label set as Switchboard-Affect (SWB-Affect). In this work, we present our approach in detail, including the definitions provided to annotators and an analysis of the lexical and paralinguistic cues that may have played a role in their perception. In addition, we evaluate state-of-the-art SER models, and we find variable performance across the emotion categories with especially poor generalization for anger . These findings underscore the importance of evaluation with datasets that capture natural affective variations in speech. We release the labels for SWB-Affect to enable further analysis in this domain. Speech emotion recognition (SER) has the potential to enhance human-computer interaction, improve our ability to monitor mental health and well-being [1], [2], and better understand customer service, entertainment, and education experiences [3], [4].

Depression commonly co-occurs with neurodegenerative disorders like Multiple Sclerosis (MS), yet the potential of speech-based Artificial Intelligence for detecting depression in such contexts remains unexplored. This study examines the transferability of speech-based depression detection methods to people with MS (pwMS) through cross-corpus and cross-lingual analysis using English data from the general population and German data from pwMS. Our approach implements supervised machine learning models using: 1) conventional speech and language features commonly used in the field, 2) emotional dimensions derived from a Speech Emotion Recognition (SER) model, and 3) exploratory speech feature analysis. Despite limited data, our models detect depressive mood in pwMS with moderate generalisability, achieving a 66% Unweighted Average Recall (UAR) on a binary task. Feature selection further improved performance, boosting UAR to 74%. Our findings also highlight the relevant role emotional changes have as an indicator of depressive mood in both the general population and within PwMS. This study provides an initial exploration into generalising speech-based depression detection, even in the presence of co-occurring conditions, such as neurodegenerative diseases.

--While multiple emotional speech corpora exist for commonly spoken languages, there is a lack of functional datasets for smaller (spoken) languages, such as Danish. T o our knowledge, Danish Emotional Speech (DES), published in 1997, is the only other database of Danish emotional speech. We demonstrate the validity of the dataset by investigating and presenting its predictive power using speech emotion recognition (SER) models. We develop SER models for EmoT ale and the reference datasets using self-supervised speech model (SSLM) embeddings and the openSMILE feature extractor . We find the embeddings superior to the hand-crafted features. The best model achieves an unweighted average recall (UAR) of 64.1% on the EmoT ale corpus using leave-one-speaker-out cross-validation, comparable to the performance on DES. Speech signals are rich in information, both linguistic (in the form of sentences and words) and paralinguistic (denoting mood and affective state). Speech also carries information about multiple, potentially personal traits of the speaker, such as age, gender, and nationality. Multiple psychological and neuroscientific models of the mind hypothesize that language and emotion are certainly linked [1]. For example, some cultures express anger more vocally, while others might be more restrained.

Voice interfaces integral to the human-computer interaction systems can benefit from speech emotion recognition (SER) to customize responses based on user emotions. Since humans convey emotions through multi-modal audio-visual cues, developing SER systems using both the modalities is beneficial. However, collecting a vast amount of labeled data for their development is expensive. This paper proposes a knowledge distillation framework called LightweightSER (LiSER) that leverages unlabeled audio-visual data for SER, using large teacher models built on advanced speech and face representation models. LiSER transfers knowledge regarding speech emotions and facial expressions from the teacher models to lightweight student models. Experiments conducted on two benchmark datasets, RAVDESS and CREMA-D, demonstrate that LiSER can reduce the dependence on extensive labeled datasets for SER tasks.

Speech emotion recognition (SER) is a vital component in various everyday applications. Cross-corpus SER models are increasingly recognized for their ability to generalize performance. However, concerns arise regarding fairness across demographics in diverse corpora. Existing fairness research often focuses solely on corpus-specific fairness, neglecting its generalizability in cross-corpus scenarios. Our study focuses on this underexplored area, examining the gender fairness generalizability in cross-corpus SER scenarios. We emphasize that the performance of cross-corpus SER models and their fairness are two distinct considerations. Moreover, we propose the approach of a combined fairness adaptation mechanism to enhance gender fairness in the SER transfer learning tasks by addressing both source and target genders. Our findings bring one of the first insights into the generalizability of gender fairness in cross-corpus SER systems.

Developing a robust speech emotion recognition (SER) system in noisy conditions faces challenges posed by different noise properties. Most previous studies have not considered the impact of human speech noise, thus limiting the application scope of SER. In this paper, we propose a novel two-stage framework for the problem by cascading target speaker extraction (TSE) method and SER. We first train a TSE model to extract the speech of target speaker from a mixture. Then, in the second stage, we utilize the extracted speech for SER training. Additionally, we explore a joint training of TSE and SER models in the second stage. Our developed system achieves a 14.33% improvement in unweighted accuracy (UA) compared to a baseline without using TSE method, demonstrating the effectiveness of our framework in mitigating the impact of human speech noise. Moreover, we conduct experiments considering speaker gender, showing that our framework performs particularly well in different-gender mixture.

In the field of large language model (LLM)-based proof generation, despite being trained on extensive corpora such as OpenWebMath and Arxiv, these models still exhibit only modest performance on proving tasks of moderate difficulty. We believe that this is partly due to the suboptimal order of each proof data used in training. Published proofs often follow a purely logical order, where each step logically proceeds from the previous steps based on the deductive rules. However, this order aims to facilitate the verification of the proof's soundness, rather than to help people and models learn the discovery process of the proof. In proof generation, we argue that the optimal order for one training data sample occurs when the relevant intermediate supervision for a particular proof step in the proof is always positioned to the left of that proof step. We call such order the intuitively sequential order. We validate our claims using two tasks: intuitionistic propositional logic theorem-proving and digit multiplication. Our experiments verify the order effect and provide support for our explanations. We demonstrate that training is most effective when the proof is in the intuitively sequential order. Moreover, the order effect and the performance gap between models trained on different data orders are substantial -- with an 11 percent improvement in proof success rate observed in the propositional logic theorem-proving task, between models trained on the optimal order compared to the worst order.

Speech emotion recognition (SER) systems often struggle in real-world environments, where ambient noise severely degrades their performance. This paper explores a novel approach that exploits prior knowledge of testing environments to maximize SER performance under noisy conditions. To address this task, we propose a text-guided, environment-aware training where an SER model is trained with contaminated speech samples and their paired noise description. We use a pre-trained text encoder to extract the text-based environment embedding and then fuse it to a transformer-based SER model during training and inference. We demonstrate the effectiveness of our approach through our experiment with the MSP-Podcast corpus and real-world additive noise samples collected from the Freesound repository. Our experiment indicates that the text-based environment descriptions processed by a large language model (LLM) produce representations that improve the noise-robustness of the SER system. In addition, our proposed approach with an LLM yields better performance than our environment-agnostic baselines, especially in low signal-to-noise ratio (SNR) conditions. When testing at -5dB SNR level, our proposed method shows better performance than our best baseline model by 31.8 % (arousal), 23.5% (dominance), and 9.5% (valence).

Natural databases for Speech Emotion Recognition (SER) are scarce and often rely on staged scenarios, such as films or television shows, limiting their application in real-world contexts. We developed and publicly released the Emotional Voice Messages (EMOVOME) database, including 999 voice messages from real conversations of 100 Spanish speakers on a messaging app, labeled in continuous and discrete emotions by expert and non-expert annotators. We evaluated speaker-independent SER models using a standard set of acoustic features and transformer-based models. We compared the results with reference databases including acted and elicited speech, and analyzed the influence of annotators and gender fairness. The pre-trained UniSpeech-SAT-Large model achieved the highest results, 61.64% and 55.57% Unweighted Accuracy (UA) for 3-class valence and arousal prediction respectively on EMOVOME, a 10% improvement over baseline models. For the emotion categories, 42.58% UA was obtained. EMOVOME performed lower than the acted RAVDESS database. The elicited IEMOCAP database also outperformed EMOVOME in predicting emotion categories, while similar results were obtained in valence and arousal. EMOVOME outcomes varied with annotator labels, showing better results and fairness when combining expert and non-expert annotations. This study highlights the gap between staged and real-life scenarios, supporting further advancements in recognizing genuine emotions.