Timestamp - Githubissues

k2-fsa / icefall

https://k2-fsa.github.io/icefall/

Apache License 2.0

841 stars 273 forks source link

Timestamp #763

Closed fightingtang closed 3 months ago

fightingtang commented 1 year ago

Is there any way to get timestamp,not only the start time of word,but also duration

csukuangfj commented 1 year ago

For CTC, yes, it is possible to get the start time and end time of a token (You have to infer the start and end time for a word from the corresponding tokens).

For transducer-based models, we can only get the start time, not the end time (so it is not possible to get the duration).

fightingtang commented 1 year ago

thank you,I will try in CTC

yaozengwei commented 1 year ago

We have already some recipes that supports getting timestamps (only the start time of word currently). E.g.,

For conformer-ctc model, please refer to https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/conformer_ctc3/decode.py.
For conformer-transducer model, please refer to https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/pruned_transducer_stateless4/decode.py