pyometrics 0.1.1

A library for creating and comparing voice biometrics using GMM models.

pyometrics

A Python library for creating and comparing voice biometrics using Gaussian Mixture Models (GMMs). You can train a biometric model from a set of audio files and then compare new samples against that model to get a similarity score.

Features

• MFCC + Delta Feature Extraction
• GMM Training via scikit-learn
• User-Defined Minimum Required Audio Files (min_files)
• Supports .wav, .mp3, .m4a, .flac by default

Installation

Install from PyPI:

pip install pyometrics

Usage

1. Train (or create) a voice biometric:

from pyometrics import create_voice_biometric

create_voice_biometric(
    folder_path="path/to/audio/folder",
    model_path="my_biometric.gmm",
    min_files=5  # specify how many audio files must exist
)

• The library expects at least min_files audio files in the folder.
• The GMM model is saved to the file my_biometric.gmm.

2. Compare a new audio sample:

from pyometrics import compare_voice_biometric

score = compare_voice_biometric(
    audio_path="path/to/new_audio.mp3",
    model_path="my_biometric.gmm"
)
print("Similarity Score (log-likelihood):", score)

• A higher (less negative) log-likelihood indicates closer similarity.

Example

Suppose you have at least 10 .mp3 files of the same speaker in samples_for_biometric/. You can do:

from pyometrics import create_voice_biometric, compare_voice_biometric

create_voice_biometric(
    folder_path="samples_for_biometric",
    model_path="example.gmm",
    min_files=10
)

score = compare_voice_biometric(
    audio_path="test_sample.m4a",
    model_path="example.gmm"
)
print("Score:", score)  # The higher, the more similar

Requirements

• Python 3.6+
• librosa
• numpy
• scikit-learn

License

This project is licensed under the MIT License.