ADAU144x : Voice Activity Detector Documentation

Hi Natan,

Sorry for the long delay. I was able to finally track down this info. Hopefully this will make it into the help file for the next release of SigmaStudio.

VAD Blocks

Voice Activity Detector w/ Accelerator takes an input signal and tracks the signal level compared to the noise floor. The Modulation Index is then later used to determine if speech is present or not

Input:

Speech Signal (Rx or Tx)

Output:

EnvelopeOut: Smoothed envelope of input signal

TopTracker: Signal tracker of the peak of the EnvelopeOut

BottomTracker: Signal tracker of the noise floor of the EnvelopeOut

ModulationIndex: Modulation level of input signal (TopTracker – BottomTracker)

Parameter Control:

LPF Smoother: 1st order LPF frequency cutoff in Hz for smoothing of input signal

Top Attack: dB/sec attack time constant value for the TopTracker

Top Release: dB/sec release time constant value for the TopTracker

Bottom Attack: dB/sec attack time constant value for the BottomTracker

Bottom Release: dB/sec release time constant value for the BottomTracker

Top Wait: Time in seconds to wait for TopTracker to be in release mode, before accelerating/increasing the release time

Top Factor: Scalar used to accelerate the release time for TopTracker

Bottom Wait: Time in seconds to wait for BottomTracker to be in attack mode, before accelerating/increasing the attack time

Bottom Factor: Scalar used to accelerate the attack time for BottomTracker

Description:

The VAD w/ Accelerator provides extra conditions to accelerate or increase the time constants in voice detection. Either the Release time for TopTracker is affected or the Attack time for BottomTracker is affected when their corresponding conditions are met. Thus, if TopTracker has been in release mode for Top Wait time, then the Top Release time constant will be increased by a scalar of Top Factor. And if BottomTracker has been in attack mode for Bottom Wait time, then the Bottom Attack time constant will be increased by a scalar of Bottom Factor.

VAD Flag takes the Modulation Index and compares it to a threshold value. If the threshold value is met for the designated amount of time, the output flag is set high, otherwise it is low.

Input:

ModulationIndex

Output

ConditionFlag: 0 or 1 output determining if speech is present or not

Parameter Control:

Threshold: dB value for comparison with Modulation Index

Count: Time value in (ms) for condition to be met before setting flag high

Thank you for getting that information. I can't find the "VAD Flag" block in my sigmastudio is it in the current install?

It seems that block did not make it in to the current release of SigmaStudio. I will try to implement a workaround using blocks already available in the library and I will post the project here when I'm finished.

How do we implement the VAD without the VAD Flag. I have tried several different ways with no success. The function has to recognize voice, not noise or music. I'm guessing the VAD Flag discerns if the input audio is voice or not.

I will be on vacation Monday the 14th and Tuesday the 15th. I won't be able to participate in EngineerZone discussions until I return on Wenesday the 16th. I will continue to support your request at that time.

Hello,

are there some new progress about the VAD-Flag or the Voice Activity Detector Documentation?

BrettG, any possibility of the VAD flag being implemented into the software? I spoke to you on the phone about a month ago regarding this issue. I have been unsuccessful with the Voice Activity Detector. It looks to me that the VAD Flag is an essential component to making this work. Please Advise.

It looks like this did not make it into the 3.4 release build. I'm sorry to make you wait. I'll push the software team to try to get this implemented in the next beta.