search

sanchit-gandhi / seq2seq-speech

Repository for fine-tuning Transformers 🤗 based seq2seq speech models in JAX/Flax.
34 stars 6 forks source link

Add RNN-T Model & Training Script #72

Closed sanchit-gandhi closed 2 years ago

sanchit-gandhi commented 2 years ago

Adds the RNN-T BPE model from NVIDIA NeMo and a training script to train the model with HF Dataset & Trainer. Template training scripts detail the configuration to train a ContextNet ASR model, large size (~144M) with Transducer loss and sub-word encoding.

sanchit-gandhi commented 2 years ago

Moved to a standalone repository: https://github.com/sanchit-gandhi/seq2seq-speech-rnnt