Originally from https://github.com/speechbrain/speechbrain/tree/develop/recipes/TIMIT/ASR/seq2seq
This folder contains the scripts to train a seq2seq RNNN-based system using TIMIT. TIMIT is a speech dataset available from LDC: https://catalog.ldc.upenn.edu/LDC93S1
python train.py train/train.yaml
| Release | hyperparams file | Val. PER | Test PER | Model link | GPUs |
|---|---|---|---|---|---|
| 21-04-08 | train_with_wav2vec2.yaml | 7.11 | 8.04 | https://drive.google.com/drive/folders/1-IbO7hldwrRh4rwz9xAYzKeeMe57YIiq?usp=sharing | 1xV100 32GB |
!pip install speechbrain !pip install transformers !git clone https://github.com/kipmccharen/sys6016_DL_project
� .. !gdown --id '1EIfBmwiT0RF3-U81-Qu5K4J27N31BdB5' ## --output /content/speechbrain_s2s_wav2vec_ckpt.zip !unzip speechbrain_s2s_wav2vec_ckpt.zip !rm speechbrain_s2s_wav2vec_ckpt.zip
� /content/data/trainwav2vec/save/ !gdown --id '1oZunuiwhMLfwtMeKAYJwr4DMjvE1LUIN' --output label_encoder.txt
� /content/
!python sys6016_DL_project/train_with_wav2vec2.py sys6016_DL_project/hparams/train_with_wav2vec2.yaml --data_folder /content/data/ --output_folder /content/data/trainwav2vec/ --new_json /content/sys6016_DL_project/data/new_train.json