[Speex-dev] VAD Questions

[Reed Hedges]

Larry Gadallah wrote:
> On 08/06/07, Steve Kann <stevek at stevek.com> wrote:
>>
>> I'd look at the speech-to-text implementations for this -- I think CMU
>> Sphinx has done something like this.
>>
> Thanks. I had a look at their web pages, and the Sphinx software looks
> interesting, but I was unable to determine if there is a "hook" in
> their system to allow simple speech _detection_ rather than
> recognition, which is what I am looking for.

Use the cont_ad_* functions (all the audio recording stuff is actually in a 
seperate libsphinx2ad library from the speech rec stuff), e.g. cont_ad_read 
returns 0 if no sound was heard.  My understanding is that detection is not 
dynamic, i.e. at startup you would call cont_ad_calib() which records some 
seconds of sound and thence will use that as a measure of ambient "silence".


Reed