VAD - simple voice activity detection in Python

This is a simple voice activity detection (VAD) algorithm in Python. It is based on simple energy-based thresholding and is intended to be used as a simple method for detecting speech in audio files when other methods cannot be used for both privacy, performance, or other reasons.

Installation

You can install the package using pip:

pip install vad

Usage

The package can be seamlessly integrated into your Python code. The following example shows how to use the package to detect speech in an audio file:

from vad import EnergyVAD

# load audio file in "audio" variable

vad = EnergyVAD(
    sample_rate: int = 16000,
    frame_length: int = 25, # in milliseconds
    frame_shift: int = 20, # in milliseconds
    energy_threshold: float = 0.05, # you may need to adjust this value
    pre_emphasis: float = 0.95,
) # default values are used here

voice_activity = vad(audio) # returns a boolean array indicating whether a frame is speech or not

# you can also use the following method to get the audio file with only speech
# speech_signal is a numpy array of the same shape as audio
speech_signal = vad.apply_vad(audio)

Audio samples

• example.wav is a sample audio file that can be used to test the package.
• example_vad.wav is the audio file with only speech after applying the VAD algorithm.
• example_vad_2.wav is the audio file with only speech direcly extracted from the original audio file using the apply_vad method.
• vad_output.png is a plot of the voice activity detected by the VAD algorithm.
• test_vad.py is the script that was used to generate the above audio files and plot.

Known issues

• There is no additional VAD algorithm implemented in this package at the moment. It may be added in the future.

License

This project is licensed under the MIT License - see the LICENSE file for details.