-
Notifications
You must be signed in to change notification settings - Fork 34
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
How many epochs does DeepSpeech2 need to converge on LibriSpeech #6
Comments
I'm middle of training bucketing model, but cer was 0.177740, 0.144390,0.126324,0.1210941 from Epoch 1 to 4 for test-clean dataset (There is no Epoch 0 result because at Epoch 0 and about 80% of data was through I accidentally powered off the server). |
Can you post your cfg file or what is your training data? |
@Soonhwan-Kwon Thanks for your response! I checked my cfg again and realized I'm only using three bigru layers. I guess that's the reason why I'm not converging as fast as expected... |
It's a relief you found the cause. Above accuracy was from the 5 bi-gru layers but if you have sufficient equipment please try 7-layers architecture. Thank you for your feed back. |
I read the issue regarding the performance of DeepSpeech2 and noticed the CER result reported by @Soonhwan-Kwon is 0.15648 at epoch 3.
It seems really promising so I'm trying to reproduce the result. But right now I'm at epoch 5 and my validation CER (vali-clean and val-other) is still 0.3122.... So I'm wondering whether I did anything wrong or was that the intended result.
Also, the test other CER on LibriSpeech reported in DeepSpeech2 paper was 0.1325. Have you guys ever come close to this number? And if so, how many epochs do you need to get there?
Thanks in advance!
The text was updated successfully, but these errors were encountered: