Polyphonic music is mentionned here to properly define the type of signals.
This is a very complex problem (especially with music) Adam ...
Separation is by characterizing the signals as in pitch /loudness/scale/chord ....
Any frequency domain tool can be used , only catch is remember Octave vs decade is the way to go .
And wavelets are ideal
Mel-frequency cepstrum (MFCC) is commonly used for speech ...
detect, all wavelets, ant shifts, and amplitude. for decoding may be ideal neural network, may be matlab, may snns. for puriti is bether use stereo sample, and detekt phase and time shift to ident how sound is how part, if is american orchester you have bigest trouble, if is viena style orchestra (by Karajan positioning) is betther, cause violin choir is on the other side like violin solo, and celo solo is to other side like celo choir.
if in the complex set of signals , the are with different pitch, cepstral analysis is a standard technique for detection and extraction of fundamental frequency of periodic signals. But, its better to use a hybrid method from different transform domains like cepstrum, fft and wavelet. Because for real set of signals they are not perfect like laboratory models and you need to extract characteristic of each of which from a proper domain.
Actually, the best would be to have a high number of microphones for the recording (I remember a case where Technical University of Delft had over 40), and to look at the correlation matrix of all... Have fun
Perhaps the easiest way to separate such signals is by N-path filters. The example of application is a notch filter that removes sound of vuvuzele, which is rich in harmonic content, see http://www.cypress.com/?docID=34215
Notch is not the only configuration possible.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Process