Vosk Tajik Speech Recognition Model
Tajik automatic speech recognition (ASR) model for sherpa-onnx. Architecture is non-streaming Zipformer2.
Version 0.60
Word Error Rate (WER)
| Dataset | WER |
|---|---|
| FLEURS Tajik | 19.49 |
| Peacockery/tajik-asr-corpus-v3 (test split) | 34.47 |
Training Data
This model was trained on Peacockery/tajik-asr-corpus-v3.
Dataset used to train alphacep/vosk-model-tg
Evaluation results
- Test WER on FLEURS Tajiktest set self-reported19.490
- Test WER on Peacockery/tajik-asr-corpus-v3test set self-reported34.470