transformer asr trained with ctc loss
LRS2 dataset: https://aistudio.baidu.com/datasetdetail/228857
Put the dataset into the data directory.
$ ls -l data/
total 3020
drwxr-xr-x 1 owen 197121 0 6月 7 22:08 lrs2/
-rw-r--r-- 1 owen 197121 36047 6月 7 21:27 test.txt
-rw-r--r-- 1 owen 197121 1191814 6月 7 21:27 train.txt
-rw-r--r-- 1 owen 197121 28132 6月 7 22:07 val.txt
ls -l data/lrs2/
total 19460
drwxr-xr-x 1 owen 197121 0 6月 7 22:02 5535415699068794046/
drwxr-xr-x 1 owen 197121 0 6月 7 22:02 5535423430009926848/
drwxr-xr-x 1 owen 197121 0 6月 7 22:02 5535496873950688380/
drwxr-xr-x 1 owen 197121 0 6月 7 22:02 5535864093654496929/
- run process_data.py to process the dataset
- run asr.ipynb to train and evaluate