Separating the UBM and the speaker voice prints.

[rbe...@aucklanduni.ac.nz]

Hello,

I would first off like to commend Amaury Crickx and the Recognito tool. As someone completely new to Speaker Recognition, I found the relative scarcity in beginner friendly tools/examples rather frustrating until I started using Recognito. So for that, I thank you for essentially introducing me to the topic.

Now to the point. I have noticed the only way to separate the UBM from the speakers you want to identify against is to set the UBM (which can be created offline, saved and reloaded) and then create VoicePrints for each speaker (which is saved to the ConcurrentHashMap 'store') at run-time. While this is a perfectly normal way to perform identification, it does mean that every time identification is performed, the VoicePrint for each speaker has to be recreated.

In an effort to setup both the UM and the speaker at the same 'offline' stage. I altered Recognito.java to make the 'store' HashMap publicly accessible, which allows me to save a serialized HashMap of previously enrolled speakers then load them every time I wish to perform identification. 

My question is, is making the 'store' HashMap publicly accessible a viable way to set the speakers without having to create their VoicePrints every time I want perform identification? I have a feeling I might doing something wrong if I am altering the original code to get the functionality I'm looking for.

Any help is much appreciated! Once again thank you for this amazing tool.

Sincerely,

Reed Bell.