TTSLab – Run and compare TTS & STT models entirely in your browser

Summary: TTSLab enables running and comparing text-to-speech and speech-to-text models directly in the browser using WebGPU and WASM without servers or API keys. Models like Kokoro 82M, Whisper, and Moonshine are downloaded once and cached locally, ensuring data never leaves the device. It supports side-by-side voice comparison, benchmarking, and an on-device Voice Agent demo.

What it does

TTSLab runs TTS and STT models fully on-device via WebGPU and WASM, allowing users to test, compare, and benchmark multiple models without internet or server dependencies.

Who it's for

It is designed for developers, researchers, and product teams evaluating speech models.

Why it matters

It eliminates the need for multiple APIs, server dependencies, and data sharing, simplifying TTS and STT model evaluation.