How we built a real-time, client-side noise suppression library without server dependencies | Datadog

This post details the development of dtln-rs, an open-source noise suppression library. It explains the DTLN algorithm, its use of STFT and LSTM, and the challenges of integrating it into WebRTC. The implementation leverages TensorFlow Lite for embedding and Rust for portability, producing WebAssembly, Node.js native modules, and native Rust targets. The library achieves real-time processing (1 second of audio in 33 ms on an M1 MacBook Pro) without server dependencies, addressing the limitations of off-the-shelf solutions and costly cloud-based services.