Universal Model

[Alysia H]

Dude, thank you sooo much for this java implementation of speaker recognition!!!

It's very easy to use and extend!

I'm using it on an android device to recognize its owner. 

You said that increasing the number of voice samples in the universal model increases the quality of the likelihood value.

For the task of verification, is it okay to have multiple voice prints of the same user(to be verified) in the universal model?

Or does this negatively affect the likelihood value?

Thanks, 

Aly

[Amaury Crickx]

Hi Aly

Thanks a lot for your kind words, they go straight to my heart!

The idea of the "universal model" is to represent an average of all possible voices so we can calculate a relative distance and provide a ratio, so yes, one need to use as many different voices as possible, using a single voice is counter productive.

Provided you have plenty of them, in order to reduce the number of false positives while verifying, you may want to use a subset of all the voices available and select the voices that are closer to the one you're identifying. E.g. grouping males and females and creating a UM for each, though some men sound like women and the other way round so I'd hand pick an initial group and measure the distance programmatically to be able to tell which group any new voice belongs to. There can be more groups than 2 (groups are actually called "cohorts" in speech terminology).

Something else, I hope to be improving the algorithm so keep your original samples around so you can recreate the voice prints using the new algo when it's out :-)

HTH

Cheers

Amaury

[Galih Reksa]

hello Aly. I really interest with your project. Use this recognito library on Android device. Is it any source of that? I want to develope it for my class experiment project.

Thanks

[Galih Reksa]

How to use universalmodel method?
I don't have any idea about that.
Could you please add the source code here.

[Amaury Crickx]

Recognito creates the universal model for you when you use the createVoicePrint and mergeVoicePrint methods.

So you don't have much to do except feeding recognito with its preferred biscuits: identified voice samples of as many people you can get hold of :-)

Do you have more specific requirements? It might help in providing an answer ;-)

Cheers

Amaury

[Urvi gola]

Hi, could you help me with using Recognito Lib into my android project? How to import?