wenet
wenet copied to clipboard
Can I get timestamp info by GPU inference?
I found timestamp info from here . But it's only for cpu. Is it possible to get timestamp from GPU inference? For example by using your docker server
@yuekaizhang is it possible?
@yuekaizhang is it possible?
Yes, it's possible to add timestamp. Currently gpu inference using this ctc_decoder, which needs to modify to add timestamps. Or we could replace it with https://pytorch.org/audio/main/models.decoder.html, which has already supported timestamps.
I could implement it next week.
@Mddct would wenet python bindings give an python API for ctc_prefix_beam_search?
@yuekaizhang is it possible?
Yes, it's possible to add timestamp. Currently gpu inference using this ctc_decoder, which needs to modify to add timestamps. Or we could replace it with https://pytorch.org/audio/main/models.decoder.html, which has already supported timestamps.
I could implement it next week.
@Mddct would wenet python bindings give an python API for ctc_prefix_beam_search?
@yuekaizhang python binding is for the convenience of using wenet。If we can binding wenet ctc prefix search in other repo to replace ctcdecoder?I've implemented a stateless binding before which maybe helpful https://github.com/Mddct/losses/blob/main/src/ctc_decoder.h
But If we introduce this api will complicate binding
@robin1001
Binding is just for CPU and it is too heavy for the task. I prefer to use torchaudio built-in decoder.
@yuekaizhang Plz note this issue: https://github.com/parlance/ctcdecode/issues/148
seems solved, close this issue