TasaractDavut Emre Taşar
Go back
Model

Wav2Vec2 XLS-R · Turkish

Wetin e dey do

Dis model dey turn Turkish speech into text with one XLS-R speech encoder wey get 300 million parameters. Dem fine-tune am with Turkish recordings from Mozilla Common Voice 8.0. One decoding step wey dem call CTC dey join repeated predictions wey follow each other and remove blank symbols to get di transcript. Di model card report word error rate of 0.2869 for di evaluation set, about 28.7%.

Original documents

Original source · huggingface.co Meta Wav2Vec2 XLS-R 300M Sequence Modeling with CTC
How to cite am

Wav2Vec2 XLS-R · Turkish · https://huggingface.co/emre/wav2vec2-large-xls-r-300m-tr