I am trying to build a model for speaker identification, and I understand that the first step is to extract the features from the audio signals that are in my database. I have done quite a bit of research and cant find how to do this extraction and to which features. How do I select which features to extract to? what would be the algorithm? Can someone please point me to either an algorithm that I could implement or a sample matlab code?
For speaker recognition, the features you should probably start with are MFCC.
There are libraries offering MFCC extraction modules, such as YAAFE, aubio (C/C++), the MIR toolbox or Dan Ellis' implementation (Matlab) - and of course speech recognition frameworks (Sphinx, HTK) provides MFCC extraction tools.
Please refer to the link.
It computes various features
- Energy Entropy Standard Deviation (std)
- Signal Energy Std by Mean (average) Ratio
- Zero Crossing Rate Std
- Spectral Rolloff Std
- Spectral Centroid Std
- Spectral Flux Std by Mean Ratio