Does the DeepSpeech v0.6.0 checkpoint is trained on whole Common Voice English Training Data or We need to train on it.
I’ll let you read and answer yourself to that question: https://github.com/mozilla/DeepSpeech/releases/tag/v0.6.0
Thank you sir for your response.
According to my understanding v0.6.0 model was trained on American English which achieves an 7.5% word error rate on the LibriSpeech clean test corpus. By that mean the checkpoint is also based on LibriSpeech not Common Voice data set. So we have to train it on Common Voice. Am i right?
No:
train_filesFisher, LibriSpeech, Switchboard, Common Voice English, and approximately 1700 hours of transcribed WAMU (NPR) radio shows explicitly licensed to use as training corpora.
1 Like