ASR-huggingface-PTBR

Creating a Wav2Vec2 model for PT-Br

Data

1. Mozilla Common Voice - PT-Br:
2. 134 hours of validated audio and transcriptions
3. 2.453 voices

Model

The idea was to use a Wav2Vec2 model and finetune it.

All the steps will be placed on the notebook, to achieve a Wer -> 0.0141 others technincs were used. The model, results and the API can be found on the HuggingFaceREPO - Victor

On the HF-Repo you can try by yourself loading a pre-recorded audio or using the Realtime Speech Recognition! Have Fun :D