Automatic Speech Recognition on Telugu Medical Domain Database
15.8WERHybrid Learnable + KenLM
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid Learnable + KenLMBackbone=Wav2Vec2, Evaluation Protocol=Fine-tuned2026.04 | 15.8 | |
| Hybrid Static + KenLMBackbone=Wav2Vec22026.04 | 17.8 | |
| Hybrid Learnable + mT5Backbone=Wav2Vec2, Evaluation Protocol=Fine-tuned2026.04 | 17.9 | |
| Hybrid Learnable + IndicBARTBackbone=Wav2Vec2, Evaluation Protocol=Fine-tuned2026.04 | 18.1 | |
| Hybrid Learnable ConfidenceBackbone=Wav2Vec2, Evaluation Protocol=Fine-tuned2026.04 | 18.9 | |
| Hybrid Static + mT5Backbone=Wav2Vec22026.04 | 19.3 | |
| Hybrid Static + IndicBARTBackbone=Wav2Vec22026.04 | 20 | |
| Hybrid Static ConfidenceBackbone=Wav2Vec22026.04 | 20.2 | |
| Baseline + KenLMBackbone=Wav2Vec22026.04 | 22.4 | |
| Baseline (without confidence)Backbone=Wav2Vec22026.04 | 24.3 | |
| Hybrid Static + IndicBARTBackbone=Whisper2026.04 | 24.8 | |
| Hybrid Static + mT5Backbone=Whisper2026.04 | 25.03 | |
| Baseline (without confidence)Backbone=Whisper2026.04 | 25.8 | |
| Hybrid Static ConfidenceBackbone=Whisper2026.04 | 26 |