search

Kitt-AI / snowboy

Future versions with model training module will be maintained through a forked version here: https://github.com/seasalt-ai/snowboy
Other
3.1k stars 999 forks source link

Feature Request: Timestamp for start / end of hotword detection #83

Open arilotter opened 7 years ago

arilotter commented 7 years ago

If the Snowboy API returned a timestamp representing the millisecond the hotword finished, systems that take other voice input after the hotword could be modified to seamlessly recognise audio said after the hotword but before Snowboy triggers.

chenguoguo commented 7 years ago

I'll keep this as a long term plan. In the meanwhile, you can try to estimate the length of the trigger words, which should give you a good estimate of the timestamp.

evancohen commented 7 years ago

+1

This would be super useful for sonus. I've tried to prototype a buffer that uses a flag for when the hotword is detected, but I've had issues getting it to work properly because all I can do is estimate at what point in the audio stream the hotword is actually detected.

astrung commented 5 years ago

This features will be helpful for so many projects, including mine, so i hope that this features will be released. It has been 3 years from last answer, so can anyone release it in near future ? Or please update its developed process ?