search

mailong25 / self-supervised-speech-recognition

speech to text with self-supervised learning based on wav2vec 2.0 framework
379 stars 114 forks source link

How to skip the lm model, i want to see the directly model prediction after ctc decoding. #42

Closed vigneshgig closed 3 years ago

vigneshgig commented 3 years ago

Hi, @mailong25 How to skip the lm model. I want to see direct model prediction after CTC decoding and also I'm getting valid_wer as 100 for every epoch, but the training loss is decreasing. why so? Note: apart from your config I just added the memory_efficient_fp16 = True, to reduce GPU memory consumption, is this affecting the wer. Please help me out. Thanks hydra_train.log

vigneshgig commented 3 years ago

Hi, @mailong25 How to skip the lm model. I want to see direct model prediction after CTC decoding and also I'm getting valid_wer as 100 for every epoch, but the training loss is decreasing. why so? Note: apart from your config I just added the memory_efficient_fp16 = True, to reduce GPU memory consumption, is this affecting the wer. Please help me out. Thanks hydra_train.log

Hi , I installed latest fairseq and pytorch 1.7.0 with cuda 10.2 in python 3.7.9 version. Now the valid_wer is decreasing and also it starts from 83~ . and also trained loss start from 54.0 comparing older version which starts from 117~ Uploading hydra_train.log… Please tell, how can i check the predicted text before the lm decoding.

mailong25 commented 3 years ago

Sorry for the late response, In the transcriber, you can set lm_type = 'viterbi' and see what happened, I haven't check if it working properly or not