finetuned-wav2vec2-960h
This model was trained as a part of my GSoC'21 (Google Summer of Code) project. It is fine-tuned on 960h of LibriSpeech dataset (train-clean-100, train-clean-360, train-other-500) and evaluated on test-clean data.
| WER (word error rate) | 5.67 |
|---|
You can find code for training here: https://github.com/vasudevgupta7/gsoc-wav2vec2.