Forums

Companies with VAD/VOX solutions?

Started by Matt Roos January 15, 2006
Hi, All.  I need to find out all I can about voice activity detection
(VAD) and companies that supply hardware or software implementations.
This could include C code for PCs or DSPs as will as hardware black
boxes with a mic/audio input and some sort of output switch or
interrupt.

I'm especially interested in non-standard solutions (e.g., not G.729,
G.723.1, AMR) since there seem to be plenty of sources for that code
(then again, I'd like to know of reputable companies with reliable
implementations).  I'm also more interested in low SNR,
high-interference environments.  This could certainly include solutions
that have been developed in the past few years (e.g., algorithms based
on HOS).

At this point I'm not interested in speech compression, just detection.

Oh, and any references to comparative studies between VADs would be
helpful too.

Thanks!
Matt

Matt Roos wrote:
> Hi, All. I need to find out all I can about voice activity detection > (VAD) and companies that supply hardware or software implementations. > This could include C code for PCs or DSPs as will as hardware black > boxes with a mic/audio input and some sort of output switch or > interrupt. > > I'm especially interested in non-standard solutions (e.g., not G.729, > G.723.1, AMR) since there seem to be plenty of sources for that code > (then again, I'd like to know of reputable companies with reliable > implementations). I'm also more interested in low SNR, > high-interference environments. This could certainly include solutions > that have been developed in the past few years (e.g., algorithms based > on HOS). > > At this point I'm not interested in speech compression, just detection. > > Oh, and any references to comparative studies between VADs would be > helpful too. > > Thanks! > Matt
VAD is a small component of speech compression/decompression algorithms. It simply calculates the energy in a 20ms-30ms speech frame and declares the speech frame either voiced, unvoice or silent. you can write a smally program using Matlab and run a recorded speech file. The energy can be calculated by using autocorrelation or cepstral methods or a combination of both. Regards, Gold
Thanks, but I'm aware of what a VAD is and some of the algorithms the
popular ones are based uponI need to do a market survey on commercially
available VADs (and non-commercial, academic versions) and their
trade-offs.  In general, VADs based solely upon energy threshold
techniques perform poorly in noisy environments--which is the scenario
I'm primarily concerned with.

Best,
Matt


Matt Roos wrote:

> Thanks, but I'm aware of what a VAD is and some of the algorithms the > popular ones are based uponI need to do a market survey on commercially > available VADs (and non-commercial, academic versions) and their > trade-offs. In general, VADs based solely upon energy threshold > techniques perform poorly in noisy environments--which is the scenario > I'm primarily concerned with. >
I don't think you will find any standalone VAD solutions. The most advanced VADs I have seen are included in the speech compression algorithms, which you have overlooked already. They are based on the assumption that the speech is non-stationary signal and the energy thresholding also. There is no doubt that the better VAD algorithm can be developed, depending on the processing power that you have and the time required to make the VAD decision. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com