| language: en | |
| tags: | |
| - speechbrain | |
| - emotion-recognition | |
| - wav2vec2 | |
| - audio-classification | |
| - pytorch | |
| license: apache-2.0 | |
| datasets: | |
| - iemocap | |
| metrics: | |
| - accuracy | |
| model-index: | |
| - name: ecapa-tdnn-emotion | |
| results: | |
| - task: | |
| type: audio-classification | |
| name: Emotion Recognition | |
| dataset: | |
| name: IEMOCAP | |
| type: iemocap | |
| metrics: | |
| - type: accuracy | |
| value: 79.1 | |
| # ECAPA-TDNN Emotion Recognition | |
| Emotion recognition model trained on IEMOCAP using ECAPA-TDNN architecture with wav2vec2 features. | |
| ## Usage | |
| ```python | |
| from speechbrain.inference.interfaces import foreign_class | |
| classifier = foreign_class( | |
| source="guyfeinberg/ecapa-emotion-v2", | |
| pymodule_file="payload.py", | |
| classname="Canary", | |
| ) | |
| ``` | |
| ## Training Details | |
| - **Architecture**: ECAPA-TDNN | |
| - **Features**: wav2vec2 + Fbank (80 mel bins) | |
| - **Dataset**: IEMOCAP (4 classes: neutral, happy, sad, angry) | |
| - **Epochs**: 30 | |
| - **Optimizer**: Adam (lr=0.001) | |
| ## Citation | |
| Research use only — Oligo Security. | |