[P] fast-vad: a very fast voice activity detector in Rust with Python bindings.
A new Rust-based VAD claims to be the fastest open-source option, with Python bindings for easy integration.
Developer AtharvBhat has launched 'fast-vad,' a new open-source voice activity detector (VAD) engineered for maximum performance. Built from the ground up in Rust and featuring Python bindings, the tool is designed to match the quality of existing VADs while prioritizing speed, simple integration, and robust streaming support. The creator states it is likely the fastest open-source VAD currently available, making it a compelling option for real-time audio processing applications where latency is critical.
Under the hood, fast-vad employs a streamlined logistic regression model that operates on frame-based audio features to maintain its high speed. It was trained using the libriVAD dataset. The project offers both batch and stateful streaming APIs, includes built-in modes with sensible defaults for ease of use, and provides configurable lower-level parameters for developers who need to fine-tune its behavior for specific use cases.
- Built in Rust with Python bindings for high performance and easy integration into Python stacks.
- Offers both batch processing and stateful streaming APIs for real-time audio pipeline support.
- Uses a fast logistic regression model on frame features, trained on the libriVAD dataset.
Why It Matters
Enables developers to build more responsive, real-time voice applications like transcription, conferencing, and assistants with lower latency.