speaker_recognition_GMM_UBM

A speaker recognition system which uses GMM-UBM for use in an Android application which helps in monitoring patients suffering from Schizophrenia.

Installing dependencies

To install all the dependencies for this project, run the following command,

pip3 install -r requirements.txt

Extracing MFCC from audio

To extract MFCC coefficients from audio samples, put all the audio files in a seperate folder and run the following command,

python3 src/speaker_recognition/extract_mfcc_coefficients.py
--audio_folder <path to the folder which contains audio>
--csv_file_name <name of the csv file that will be created>
--opt combined

Creating Universal Background Model

To run UBM training run the following code,

python3 src/speaker_recognition/speaker_recognition.py 
--csv_file <path to MFCC coefficients file> 
--operation ubm

Map adaptation using the created GMM-UBM model

To run MAP adaptation,

python3 src/speaker_recognition/speaker_recognition.py 
--csv_file <path to MFCC coefficients file> 
--operation map 
--ubm_file <path to the ubm file created after GMM-UBM model creation>

For testing the map adapted model,

python3 src/speaker_recognition/testing_model.py
--map_file_name <path to map adapted .npy file>
--ubm_file_name <path to ubm .npy file>
--test_csv_file <path to the csv file of test speaker>
--N 1500