Who speaks?

Speaker recognition using Mel Frequency Cepstral Coefficients (MFCC) and Linde-Buzo-Gray (LBG) clustering algorithm. The implementation is based on this matlab tutorial. It gives about ~65% of correct results using this data set.

Prepare audio files

bash scripts/preprocess-audio.sh

Creates audio-preprocessed folder

Train

./src/cli.js train ./audio-preprocessed/train/

Creates training-results.json

Recognize

Requires training-results.json

./src/cli.js recognize ./audio-preprocessed/test/F101_test_1.wav

Test performance