I heard that i can use FFT for detecting sound activities. For example if i have sound stream i want to get only those chunks where there is some activity is going on and leave out those that are just silence
Posted on 2003-07-28 18:48:33 by masnick[CCCP]
Yes you could use this.

FFT takes a stream of data, and separates the different frequencies, and puts them into "bins". Each bin represents a frequency, and the number in the bin represents the "strength" in signal from that frequency.

Typical noise, is uniform in "strenght" and will randomly affect all bins. So you could:

1) Trap a frequency range common to human voices (Min bin, max bin)
2) Trap for signal strengths in these bins higher than you "noise threshold" strenght.

This works on the assumption that the speakers voice will create at least one frequency bin signal strength that is higher than your threshold.

Hope this helps.. (I wrote a FFT algo a while back for MASM here... )

Posted on 2003-07-28 18:56:28 by NaN
Thanx. I guess i'll need to do some additional searching and reading.
Posted on 2003-07-28 18:59:04 by masnick[CCCP]