Speaker Dependent voice recognition on a microcontroller?


I'm doing a project which would take voice input from only certain people (say 2-3). But I'm unable to find how to specify these voices. How do I distinguish the voices. What property of voice makes it different from others which can be quantized?

K. Nik

Posted 2016-12-11T07:29:40.067

Reputation: 11

Question was closed 2017-01-15T16:40:25.647

I'm voting to close this question as off-topic because it is not about sound design, but AI design. – Tetsujin – 2016-12-11T09:08:33.770

1I don't think what you are asking for is possible within the current bounds of sound engineering. Speech recognition is limited the fact that it interprets frequencies and not things like timber which are not easily measured. – None – 2016-12-12T18:47:21.007

I'm voting to close this question as off-topic because this is more suited to the dsp or ai forums. Probably needs to be asked in both actually. – Mark – 2016-12-18T11:15:24.330



Do you want to adjust settings on speaker according to voice's characteristics automatically? You then need to extract mel cepstral frequency coefficients and also you need a database with voices for training, but this answer is off-topic here. Any way, you can use a program named praat and mark in .wav file (for example) and then extract coefficients you want to work with. It's not simple. you will need programming and sound classification knowledge for that but it is pretty easy to find help online. Just search for sound classification or voice classification. Sorry for bad english.

Mike Kampitakis

Posted 2016-12-11T07:29:40.067

Reputation: 31


Just re-reading this question and have had another idea. If this is a recording environment, simply put lavalier microphones on the talent that you want to record and then use a simple forward-looking gate to open the channel on the voice you want to let into the mix when there is voice on that channel - simple level-dependent gate should do it.


Posted 2016-12-11T07:29:40.067

Reputation: 7 535