Speech Recognition on FLEURS Kazakh (test)
15.4WER (Oracle)Fine-tuning Community Model (KSC2)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Fine-tuning Community Model (KSC2)Training Data=CVC + FLEURS2026.03 | 15.4 | 3.4 | 7.1 | 2 | |
| Fine-tuning Community Model (KSC2)Training Data=Songs + CVC + FLEURS2026.03 | 15.8 | 3.7 | 7.9 | 2.3 | |
| Fine-tuning Community Model (KSC2)Training Data=FLEURS only2026.03 | 16 | 3.6 | 7.3 | 2.1 | |
| Fine-tuning Community Model (KSC2)Training Data=Songs + FLEURS2026.03 | 16.3 | 4.1 | 8.4 | 2.7 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=Songs + CVC + FLEURS2026.03 | 19.5 | 3.9 | 11.8 | 2.6 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=Songs + FLEURS2026.03 | 19.7 | 3.9 | 11.8 | 2.6 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=CVC + FLEURS2026.03 | 19.7 | 4 | 11.8 | 2.6 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=FLEURS only2026.03 | 21 | 4.4 | 13.6 | 3.1 | |
| Fine-tuning Community Model (KSC2)Training Data=Songs + CVC2026.03 | 26.5 | 7.2 | 11.5 | 4.8 | |
| Fine-tuning Community Model (KSC2)Training Data=Songs only2026.03 | 32.6 | 7.9 | 11.7 | 4.7 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=Songs + CVC2026.03 | 33.8 | 7.5 | 23.7 | 5.6 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=Songs only2026.03 | 33.9 | 7.6 | 23.7 | 5.7 | |
| Fine-tuning Community Model (KSC2)Training Data=CVC only2026.03 | 35.7 | 8.3 | 10.8 | 4.5 | |
| Community fine-tuned (KSC2)Training=Baseline2026.03 | 36 | 8.8 | 11.3 | 5.1 | |
| Whisper Large-V3 TurboMode=zero-shot2026.03 | 38 | 9 | 21 | 6 | |
| Whisper Large-V3Mode=zero-shot2026.03 | 41.1 | 9.5 | 33.1 | 7.8 | |
| Fine-tuning Whisper Large-V3 TurboTraining Data=CVC only2026.03 | 48.1 | 11.3 | 37.9 | 9.2 |