Open akshat9425 opened 5 years ago
Hi akshat,
You would like to train a UBM GMM model for that. Here, we train a universal background model of lots of speaker to learn a generic acoustic space and then do discriminative training of speakers. This one is the best paper to follow.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.117.338&rep=rep1&type=pdf
@abhijeet3922 Thanks alot for your valuable reply i go through your provided doc that's great to understood physics but will you please help me that how could i implement same thing in programming world
Thanks in advance Any help is appereciated
i want to give my own dataset of voices for speaker identification and verification will you please tell on what parameters voice should be recorded i.e what should be its bit resolution, sampling rate,audio channels and normalized or not
I have to put if-else check on score value to detect and eliminate those speakers whose voice are not trained i hope you understood what i want to say
please reply thanks in advance
Read Documentation, try on VoxForge and other datasets first, then implement on your own dataset
Is there any Implementation of GMM-UBM model already available on Github, if so please let me know.
I used link: https://github.com/abhijeet3922/Speaker-identification-using-GMMs
it works well for speaker identification but while verification the voice not present in dataset is predicted as any other random voice from dataset how do i predict those voices to eliminate i need to integrate this for login and registration with voice
please reply thanks