Atahan
Newbie level 2
Hi, I'm working on feature extraction from speakers voice using MFCC. I was thinking of improving the accuracy of speaker verification by weighing the different feature parameters (12 of them) depending on their intra-speaker variability. By letting the person speak his name several times during enrollment and measuring which parameters have least intra-speaker variability. Those parameters with least intra-speaker variability will then be weighed heavier then other parameters in the authentication phase.
Does this method have a name? Are there superior alternatives to this that can be used for pattern matching?
Another question while I'm at it: What is the most practical method for pattern matching to authenticate a person by their voice when using MFCC parameters? Some sort of euclidean distance between the MFCC paramters comes to mind. If so are all the parameters "worth" equal in the sum of deviation?
Does this method have a name? Are there superior alternatives to this that can be used for pattern matching?
Another question while I'm at it: What is the most practical method for pattern matching to authenticate a person by their voice when using MFCC parameters? Some sort of euclidean distance between the MFCC paramters comes to mind. If so are all the parameters "worth" equal in the sum of deviation?