search

ubclaunchpad / minutes

:telescope: Speaker diarization via transfer learning
https://medium.com/ubc-launch-pad-software-engineering-blog/speaker-diarisation-using-transfer-learning-47ca1a1226f4
27 stars 5 forks source link

pre trained model #137

Open prashantmaheshwari94 opened 6 years ago

prashantmaheshwari94 commented 6 years ago

HI I have two questions.

  1. Is this implementation trained on any data? What was the accuracy?
  2. Is there any pre trained model available?

Cheers

grig-guz commented 6 years ago

@prashantmaheshwari94

  1. The base implementation is trained on LibriSpeech (it has a lot of sound data per speaker). The accuracy was ~95%
  2. The base model is here: https://github.com/ubclaunchpad/minutes/tree/master/minutes/models/base/cnn