How to seperate the talk of different speakers in a record

Hi, I am now working on speech signal processing and there's a problem: how to seperate the conversation between a specific person and a stranger(chang always) so that we can analyze the speech signal of the specific person or the other. Thanks for your help! Oscar

[KSSV]

You need to do fft analysis.....get the frequencies and filter out the required frequencies.

[Walter Roberson]

See also Blind Source Separation

[Lizhao Du]

Thanks for your answer. But the question is that different speakers may also have the same frequency. How to find out a feature that matches a specific person and then seperate the conversation so that the signal of only one of them can be selected? Or try to use a classifier? The truth is that using GMM model based on MFCC doesn't work, cause the signal database requires both of the speakers, which is not practical.

[Lizhao Du]

Hi, Walter, could you please send me a link about Blind Source Seperation?

[Walter Roberson]

https://www.mathworks.com/matlabcentral/answers/?term=tag%3A%22bss%22

[Lizhao Du]

Thanks for your help, Walter.

Answer 1

[Image Analyst]

See these links:

https://en.wikipedia.org/wiki/Cocktail_party_effect

http://www.cnl.salk.edu/~tony/ica.html

http://www.cis.hut.fi/projects/ica/fastica/

http://www.cis.hut.fi/projects/ica/icademo/

http://research.ics.aalto.fi/ica/cocktail/cocktail_en.cgi

You can/should use BSS/ICA to solve it.