Voice Activity Detector (VAD) block description
Hello All,
We have a Analog Devices A2B bus evaluation kit which is based on ADAU1452 and as far as I understood features also one of these programmable audio dsps.
Now we have realized a design with SigmaStudio which uses the Voice Activity Detection block (which was already available inside the SigmaStudio Library).
I use the envelope output of the VAD block and connect it to VAD Flag unit and use its output as the detected speed acitivty flag. Is this correct?
Somewhere I read I should use the modulation index output. But that output is all the time zero. The Envelope output is not zero and changes according to input audio stream. But I do not understand why the modulation index is always zero.
Where can I read the algorithm implemented in this block?
Thanks.
Audio Signal Processors
adau1452
SigmaDSP Audio Processors
Hi Mamsadegh,
I see that no one responded yet to your question. I have no experience with the 1452, only the 1761.
But the VAD on the 1761 does not seem to function very well in noisy environments or when the speech
is not of high quality.
Some more documentation can be found here:ADAU144x : Voice Activity Detector Documentation
& on the wiki: Voice activity detection - Wikipedia, the free encyclopedia
Someones implementation on a 1761 can be found here:Voice Activity Detector Implementation
KJBob made a custom VAD: VAD for ADAU170x
I made a VAD based on the Zero-Crossings Rate. This distinguishes between noise and speech.
If you combine it with a level detector you have a nice speech detector. Be sure to read the whole
post as I describe numerous improvements. It can be found here: Zero Cross Rate VAD.
As the 1452 possesses frequency domein capabilities, you can improve the VAD by looking at
the frequency domain.
Hope this helps,
Benjamin
Please log in to reply.