Sonopy

A simple audio feature extraction library

Sonopy is a lightweight Python library used to calculate the MFCCs of an audio signal. It implements the following audio vectorization functions:

• Power spectrogram
• Mel spectrogram
• Mel frequency cepstrum coefficient spectrogram

Features

• Lightweight
• Tiny, readable source code
• Visualize steps in calculation

Usage

import numpy as np
from sonopy import power_spec, mel_spec, mfcc_spec, filterbanks

sr = 16000
audio = np.random.random((2 * 16000))

powers = power_spec(audio, window_stride=(100, 50), fft_size=512)
mels = mel_spec(audio, sr, window_stride=(1600, 800), fft_size=1024, num_filt=30)
mfccs = mfcc_spec(audio, sr, window_stride=(160, 80), fft_size=512, num_filt=20, num_coeffs=13)
filters = filterbanks(16000, 20, 257)  # Probably not ever useful

powers, filters, mels, mfccs = mfcc_spec(audio, sr, return_parts=True)

Installation

pip install sonopy
pip install "sonopy[example]"  # For example.py
pip install "sonopy[comparison]"  # For comparison.py

Speed Comparison

Library links:

• Sonopy
• SpeechPy
• python_speech_features
• librosa

Credits

Thanks to SpeechPy for providing an example of the concrete calculations for MFCCs. Much of the calculations in this library take influence from it.