TTSLab

The product is a browser-based tool for testing and comparing text-to-speech and speech-to-text models that runs inference locally to avoid sending text or audio to external servers. It is for developers evaluating TTS options, researchers benchmarking model quality, product teams comparing voices, and general consumers testing speech AI. It is delivered as an open source, client-side application using WebGPU and WASM with models downloaded and cached locally, requiring no servers, API keys, or installation and keeping all data on device.

Key features

• Test TTS models in browser
• Test STT models in browser
• Run inference via WebGPU
• Run inference via WASM
• Compare models side-by-side
• Benchmark model quality
• Voice Agent preview in browser
• ONNX-native model support
• Cache models locally for reuse
• No server-side processing
• No data collection