OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Vad

Vad is a voice activity detection toolkit that pairs MATLAB-based multi-resolution cochleagram feature extraction with Python and TensorFlow classifiers, and includes a recorded dataset and a trained model.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Vad — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Vad when you need a research-oriented VAD codebase with several classifier approaches and a real-world recorded dataset, and can work with its MATLAB and Python prerequisites.

Who it is for

  • Speech-processing researchers comparing neural voice activity detection approaches.
  • Developers exploring speech-active region detection for recorded audio.
  • Researchers using a manually annotated dataset of conversational speech recorded in varied environments.

Documented uses

  • Detect speech-active regions in recorded audio.
  • Compare ACAM, bDNN, DNN and LSTM-RNN classifiers using MRCG features.
  • Explore or train voice activity detection models using the included data and configurations.
Before you choose

Requirements & limits

  • The feature extraction stage uses MATLAB; the classifiers use Python with TensorFlow.
  • The README lists Python 3, TensorFlow 1.1–3 and MATLAB 2017b as prerequisites.
  • MRCG extraction is relatively slow compared with classification.
  • For other speech data, the project specifies a 16 kHz sampling rate.
  • Post-processing settings depend on the application; the README says there is no optimal parameter set for all uses.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run the included example on the project dataset

Use the provided trained model and MATLAB example to view voice activity detection results.

Before you start

  • Python 3
  • TensorFlow 1.1–3
  • MATLAB 2017b
  • The project repository and its included model and dataset

Steps

  1. Obtain the VAD toolkit and its included files.

  2. Add the project path in MATLAB.

  3. Run the example script main.m in MATLAB.

Expected outcome

The README describes an example result produced with the trained model based on the project dataset.

Guide limitations

  • For other speech data, the README specifies a 16 kHz sampling frequency.
  • MRCG feature extraction takes relatively long compared with classification.
Documentation 1 ↗
Source-based practical guide

Prepare the documented model-training workflow

Follow the repository’s stated preparation and training stages for its VAD models.

Before you start

  • Python 3
  • TensorFlow 1.1–3
  • MATLAB 2017b
  • The sample database, model specifications and project files

Steps

  1. Review the sample database under the project data/raw path to understand its format.

  2. Review model specifications under the project configure path.

  3. Add the project path in MATLAB.

  4. Follow the documented two stages: MRCG extraction, then model training.

Expected outcome

The README identifies the training stages and model choices: ACAM, bDNN, DNN and LSTM.

Guide limitations

  • The repository describes training as a two-stage procedure; this guide does not establish that it has been run.
  • MRCG extraction is relatively slow compared with classification.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
jtkim-kaist/VAD · Official homepage title · retained in its source language
How we review sourcesSuggest a correction