TTSLab

Introduction to TTSLab

TTSLab is a browser-based platform that allows users to test and compare Text-to-Speech (TTS) and Speech-to-Text (STT) models without the need for a server, API keys, or data collection. Built using WebGPU and WASM, it enables on-device inference, ensuring that all text and audio remain within the user's device. This makes TTSLab ideal for developers, researchers, and product teams who want to evaluate model performance, compare voices, or build applications with on-device speech AI.

The platform supports a wide range of models such as Kokoro 82M, Whisper, Moonshine, and Supertonic 2. These models are downloaded once and cached locally, allowing for instant reuse. TTSLab also includes features like voice comparison, benchmarking, and a Voice Agent that runs entirely in the browser. It emphasizes privacy, speed, and open-source collaboration, making it a powerful tool for anyone working with speech AI.

Takeaways

• Run TTS & STT models directly in your browser
• No server, no API keys, no data collection
• On-device inference ensures privacy and zero latency
• Models are downloaded once and cached locally
• Supports multiple TTS and STT models including Whisper, Moonshine, and Kokoro 82M
• Includes voice comparison, benchmarking, and a Voice Agent
• Fully open source (MIT license)

How TTSLab Works

TTSLab operates through a simple four-step process:

1. Pick a Model: Browse a directory of TTS and STT models or compare them side by side.
2. Downloads Once: The selected model is downloaded and stored locally in the browser for future use.
3. Runs Locally: Inference is performed entirely in the browser using WebGPU or WASM, eliminating the need for external servers.
4. Data Stays Private: All text and audio processing occurs on the user’s device, ensuring no data leaves the browser.

Core Benefits and Applications