search

juliuskunze / speechless

Speech-to-text based on wav2letter built for transfer learning
https://arxiv.org/pdf/1706.00290.pdf
MIT License
96 stars 25 forks source link

Output only blank #4

Open wuyifat opened 6 years ago

wuyifat commented 6 years ago

Hi, I tried to training from scratch using minimal english. The output quickly converged to underscore, so all the predicted transcripts are just empty strings. Any suggestions? Thank you.

salexo commented 6 years ago

How many epochs did you wait for? For me the first one or two epochs were gibberish, the next couple were empty, and only after then the model started outputting something sensible.