Any possibility to train/finetune HuBert using custom data

facebookresearch / textlesslib

Library for Textless Spoken Language Processing

MIT License

518 stars 50 forks source link

Closed KingStorm closed 2 years ago

KingStorm commented 2 years ago

This is really great job! I am wondering if textlesslib could also support training or finetuning HuBert with custom data.

For example not just in English but other language data.

JadeCopet commented 2 years ago

Hello,

Thanks for your interest!

Training or fine-tuning of HuBERT in textlesslib is not planned at the moment. Please refer to fairseq's codebase to do so: https://github.com/pytorch/fairseq/tree/main/examples/hubert

KingStorm commented 2 years ago

Sure, thanks.