imranparuk / speaker-recognition-3d-cnn

Keras + pyTorch implimentation of "Deep Learning & 3D Convolutional Neural Networks for Speaker Verification"
MIT License
29 stars 12 forks source link

Does it accept multi-modeling at once? #3

Open MSAlghamdi opened 5 years ago

MSAlghamdi commented 5 years ago

Hi Imran,

Thanks a lot for your nice work.

Are your version either the original one able to create more than one model or verify multiple audio samples at once?

For complicated reasons, I need to make many models for many different speakers and verify a bunch of audio samples at once.

imranparuk commented 5 years ago

Hmm, from what I understand I do think its possible, just run the inference code over all the audio samples you have. Unless I am miss-understanding you.

imranparuk commented 5 years ago

I didn't add any inference code. When I get time ill get round to doing it.

MSAlghamdi commented 5 years ago

Let me ask again more clearly:

Is it a way to create many models "for different speakers (IDs)" and verify multiple audio samples "for different speakers too" at once (by putting samples for more than one speaker in one txt and audio files)?

imranparuk commented 5 years ago

I think for this implementation, what is required for you in an enrollment stage. However my implementation is just a simple one currently. I don't think ill ever set it up in the general speaker recognition format. Therefore you can check out the original authors implementation 3D-CNN, or you could change this code to a Siamese implementation which I feel is the best way.

Let me know if this helps