kensho-technologies / pyctcdecode

A fast and lightweight python-based CTC beam search decoder for speech recognition.
Apache License 2.0
421 stars 89 forks source link

Hotwords runtime is slow #109

Open nirraviv opened 1 year ago

nirraviv commented 1 year ago

Hi, thanks for the project! I'm decoding on top of NeMo model with and without variable amount of hotwords and get the following runtime:

Is there a plan on improve this amazing feature run-time using trie or any other solution?

yashjogi commented 1 year ago

Are you using LM as well during decoding? @nirraviv

nirraviv commented 1 year ago

@yashjogi yes