wav2vad is a command-line tool that performs voice activity detection (VAD) on WAV audio files. It utilizes the Silero VAD model and requires the ONNX runtime for inference. The tool processes the input audio file and outputs the detected voice activity occurences in JSON format.
- ONNX Runtime: Required for running the Silero VAD model.
- libsndfile: Required for reading WAV audio files.
-
Install ONNX Runtime:
- Follow the installation instructions provided in the ONNX Runtime repository.
-
Install libsndfile:
- You can install libsndfile using your package manager. For example, on Debian, you can use
apt
:sudo apt install libsndfile1-dev
- You can install libsndfile using your package manager. For example, on Debian, you can use
-
Download the Silero VAD Model:
- Download the Silero VAD model from here.
-
Build wav2vad:
- Clone this repository:
git clone https://github.com/eja/wav2vad.git
- Navigate to the cloned directory and build the executable:
cd wav2vad make
- Clone this repository:
After building the wav2vad
executable, you can use it to perform voice activity detection on WAV audio files.
Usage: ./wav2vad <path_of_the_onnx_model> <path_of_the_wav_file>
<path_of_the_onnx_model>
: Path to the ONNX model file for voice activity detection.<path_of_the_wav_file>
: Path to the input WAV audio file.
The output will be in JSON format and will contain the detected voice activity time along with confidence scores.
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.
- This project relies on ONNX Runtime for executing the VAD model.
- The libsndfile library is used for reading WAV audio files.
- This project is based on the Silero VAD example available at snakers4/silero-vad.