search

google / uis-rnn

This is the library for the Unbounded Interleaved-State Recurrent Neural Network (UIS-RNN) algorithm, corresponding to the paper Fully Supervised Speaker Diarization.
https://arxiv.org/abs/1810.04719
Apache License 2.0
1.55k stars 320 forks source link

Pretrained Models #2

Closed dsleo closed 5 years ago

dsleo commented 5 years ago

Hello,

I know this is exactly the model used in the paper, but was wondering if you planned to release pretrained model on the datasets of section 4.4 of the paper ?

Many thanks.

wq2012 commented 5 years ago

Sorry we cannot do that for a couple of reasons.

  1. The libraries for training speaker recognition models have heavy dependencies on Google's internal infrastructure. They cannot be open sourced.
  2. The data we used for training the d-vector models are proprietary data, thus the models generated from those data cannot be shared.
  3. The models are in an internal format that does not have public APIs.

We did lots of work to isolate the UIS-RNN algorithm and make it into open source. You are welcome to use any third-party libraries and public datasets to train your own speaker recognition model, and hook it to the UIS-RNN algorithm.