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
3.08k stars 997 forks source link

Understanding how to feed audio #240

Open snktagarwal opened 7 years ago

snktagarwal commented 7 years ago

Hi Guys,

Amazing work on the hotword detection. About time someone came up with it 👍

I had a question about how to use this properly. I used a dummy word, say "word" to train my model. Now if I have a continuous speech being said -- how should I use the APIs to feed them the speech on the trained model. I see that if I say the occurance separately i.e. just "word" then the detector detects is well.

But if the speech is "hello word hello" then I cannot get the detector to detect it. This is while using Go APIs. I expect this to work since the online demo with my model (I think?) works. What am I missing here?

chenguoguo commented 7 years ago

When you provide the "word" audio samples, you can cut the silence/noise around the actual pronunciation of "word". That might help.