Speech Emotion Recognition in Multimodal Environments with Transformer: Arabic and English Audio Datasets
Bibliographic record
Abstract
Speech Emotion Recognition (SER) is a fast-developing area of study with a primary goal of automatically identifying and analyzing the emotional states expressed in speech. Emotions are crucial in human communication as they impact the effectiveness and meaning of linguistic expressions. SER aims to create computational approaches and models to detect and interpret emotions from speech signals. One of the primary applications of SER is evident in the field of Human-Computer Interaction (HCI), where it can be used to develop interactive systems that adapt to the user's emotional state based on their voice. This paper investigates the use of speech data for speech emotion recognition. Additionally, we applied a transformation process to convert the speech data into 2D images. Subsequently, we compared the outcomes of this transformation with the original speech data, aligning the comparison with a dataset containing labeled speech samples in both Arabic and English. Our experiments compare three methods: a transformer-based model, a Vision Transformer (ViT) based model, and a wave2vec-based model. The transformer model is trained from scratch on two significant audio datasets: the Arabic Natural Audio Dataset (ANAD) and the Toronto Emotional Speech Set (TESS), while the vision transformer is evaluated alongside wave2vec as part of transfer learning. The results are impressive. The transformer model achieved remarkable accuracies of 94% and 99% on ANAD and TESS datasets, respectively. Additionally, ViT demonstrates strong capabilities, achieving accuracies of 88% and 98% on the ANAD and TESS datasets, respectively. To assess the transfer learning potential, we also explore the Wave2Vector model with fine-tuning. However, the findings suggest limited success, achieving only a 56% accuracy rate on the ANAD dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.004 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".