search

ALIZE-Speaker-Recognition / android-alize

ALIZE for the Android platform.
GNU Lesser General Public License v3.0
35 stars 17 forks source link

about the global model world.gmm #11

Closed AhmadKhattak closed 6 years ago

AhmadKhattak commented 6 years ago

Hi,

I’m trying to run the ALIZE android library for speaker verification, i’ve downloaded the library and followed the steps in https://github.com/ALIZE-Speaker-Recognition/android-alize

however, i’m confused regarding the world.gmm file, i know its the universal background model but i don’t know how to generate it.

Edit: I've used another source on the web to generate a universal background model https://appliedmachinelearning.wordpress.com/2017/06/14/voice-gender-detection-using-gmms-a-python-primer/ the male.gmm model was generated following the link. Now, i'm trying to use this male.gmm model for the world.gmm model and use the train_data samples available in the link to perform a demo of speaker verification, however i'm getting the following error,

screen shot 2018-03-17 at 2 55 53 pm

Any help would be much appreciated, Thanks !

RITCHIEHuang commented 6 years ago

@AhmadKhattak So how do you generate your word.gmm file ? image

AhmadKhattak commented 6 years ago

@RITCHIEHuang to generate a world model, what i've been doing is from http://alize.univ-avignon.fr/ downloaded Tutorial for LIA_SpkDet — GMM/UBM System and then read through those .sh files to understand how to generate the world.gmm file.

Basically,

  1. Extract MFCC features from your audio files using SPRO or HTK.
  2. Perform Feature Normalization using the LIA_RAL library methods such as NormFeat, EnergyDetector.
  3. Use the TrainWorld method from LIA_RAL library to generate your world.gmm (Universal Background Model)