silence detection

[carthia]

Hello All,
I need to determine speaking time, but without pressing any button
(such as start speaking/stop speaking), e.g. how long each
person(microphone) was speaking during one meeting, and s/he should
see results in real-time.
And I am stuck, do you have any solutions??

Thanks in advance

 

[MBR]

You can run a low-pass filter (simple exponential average,etc.) over your
samples to get something to start with.
You can have a really long window to get a baseline to use as a
descision-point to use for a faster average (with some absolute cut-offs as
well). You could combine the power measurements with speech/non-speech
detection using the front-end of a speech recognizer as well, or a
make-shift one if you feel really crafty. If you're doing speech reco
already, it's best to use liberal power cutoff's combined with the output of
the recognizer. If this is far-field (no close-talking mic.) then you have
your work cut out for you...

 

[Nicholas Paldino [.NET/C# MVP]]

carthia,

If you have access to the waveform data coming in over the device, you
will have to keep track of the amplitude of the waveform. When you feel
that the amplitude is at a level that indicates "silence" (and this is not
0, because you will almost never get that due to ambient noise and whatnot),
you start a timer which has an interval of the amount of time you want to
wait for silence. As you continue to process the waveform, if the amplitude
of the waveform goes above your threshold, then disable the timer.

Then, if your timer goes off, you know you have had "silence" for an
applicable period of time, and you can process from there.

Hope this helps.

 

[carthia]

thx for quick response,
it really helps