TTSLab
Text-to-speech that runs in the browser via WebGPU
Last verified:
What is TTSLab?
TTSLab is a browser‑based platform that lets you test and compare text‑to‑speech (TTS) and speech‑to‑text (STT) models entirely on your device, without any server‑side processing, API keys, or data collection. It runs models using WebGPU and WASM, so all text and audio stay local to your browser, enabling fast, private speech‑AI experimentation and benchmarking. The tool provides a curated directory of TTS and STT models, side‑by‑side comparisons, and a simple one‑click workflow: pick a model, download it once, run inference locally, and re‑use the cached weights later.
Key features include in‑browser TTS and STT model testing, WebGPU‑accelerated inference for low latency, and a built‑in ‘voice agent’ that chains speech‑to‑text, a language model, and text‑to‑speech into a fully local conversational AI. TTSLab also supports standardized evaluation by feeding the same input to multiple models at once and benchmarking generation speed on your own hardware, making it well suited for developers, researchers, and product teams who want to compare voice‑AI quality, latency, and privacy profiles. The project is open source under the MIT license, with all code and model‑integration logic available on GitHub.
TTSLab is designed for developers who want to vet TTS/STT models before deploying them, researchers who need reproducible, hardware‑specific benchmarks, and product teams who care about on‑device privacy and low‑latency voice experiences. Because models run client‑side, it is especially useful for scenarios involving sensitive text such as medical notes, legal content, or private conversations, where sending data to a cloud service is not acceptable. The no‑setup nature—no Python, no dependencies, no API keys—also makes it accessible to non‑engineers who want to quickly audition different voices or transcription models right in the browser.
TTSLab pricing
Pricing model: Freemium
TTSLab is a fully open‑source, MIT‑licensed tool that runs entirely in the browser with no server‑side infrastructure, so there is no paid subscription or usage‑based pricing; the platform is free to use for individuals and organizations. The only costs are local resources such as GPU memory and bandwidth when downloading model weights, but TTSLab itself does not charge for access, model serving, or private usage. Since there are no proprietary cloud tiers, there is no separate free tier or paid plan structure; all features, including model testing, side‑by‑side comparisons, and the voice agent, are available at no cost.
TTSLab pros
- Runs TTS and STT models entirely in the browser with no server
- No API keys or cloud accounts required to get started
- No data is sent to external servers; text and audio stay local
- Zero data collection or tracking from the TTSLab platform
- Models are downloaded once and cached locally for instant reuse
- Powered by WebGPU for fast, low‑latency on‑device inference
- Supports both TTS and STT models in a single unified interface
- Standardized side‑by‑side model comparisons with shared inputs
- Built‑in voice agent that chains STT → LLM → TTS in the browser
- Real‑time speed benchmarking on your own hardware
- Open‑source under the MIT license with full code transparency
- No Python or local dependencies needed to run models
- Easy primitives for evaluating model quality and latency
- Helps assess privacy and compliance implications of different voices
- Accessible from any modern browser without desktop installs
TTSLab cons
- Requires a relatively modern GPU and browser that supports WebGPU
- Model size and inference speed depend heavily on local hardware
- Some models may be slow or unusable on low‑end devices
- Browser‑only UX limits advanced scripting or batch workloads
- Fewer production‑grade deployment options compared to cloud APIs
- No built‑in team collaboration or shared workspaces
- Limited troubleshooting guidance if WebGPU or WASM fails
- Model selection is still relatively small compared to large cloud providers
Frequently asked questions about TTSLab
How does TTSLab run TTS and STT models in the browser?
TTSLab loads ONNX‑optimized TTS and STT models into the browser and runs them using WebGPU or WASM for hardware‑accelerated inference. Models are downloaded once and cached locally, so subsequent uses start immediately without re‑downloading. All text and audio are processed entirely on your device; nothing is sent to a server.
Do I need an API key or cloud account to use TTSLab?
No; TTSLab does not require any API keys, cloud accounts, or backend services. Everything runs client‑side in your browser, so you can start testing models as soon as you load the page, with no registration or authentication step.
Is my data private when using TTSLab?
Yes; your text and audio are processed only in your browser and never leave your device. TTSLab performs zero data collection or tracking, so there is no server that can see what you are transcribing or synthesizing.
Can I use TTSLab on a low‑end laptop or older PC?
You can use TTSLab on any device that supports a modern browser and WebGPU, but model‑inference speed depends heavily on your hardware. Smaller models such as Moonshine Tiny will run well on modest machines, while larger models may be slow or fail to load on low‑end systems.
What kinds of models does TTSLab support?
TTSLab focuses on ONNX‑optimized TTS and STT models that can run efficiently in the browser, such as Kokoro 82M, SpeechT5, Piper, Whisper variants, and Moonshine Tiny. The platform aims to standardize evaluation across these models so you can compare them fairly on the same inputs.
How does TTSLab handle model downloads and storage?
When you pick a model for the first time, TTSLab downloads its weights into your browser’s local storage. On subsequent visits, those weights are re‑used from the cache, so you do not need to re‑download the model each time you want to test it.
Is TTSLab open source and can I contribute models?
Yes; TTSLab is fully open source under the MIT license, with the codebase and model‑integration layer hosted on GitHub. The project welcomes contributions such as adding new TTS or STT models that have ONNX exports, improving the UI, writing documentation, or reporting bugs and feature ideas.
What is the voice agent feature?
The voice agent is an experimental TTSLab component that chains speech‑to‑text, a language model, and text‑to‑speech entirely in the browser, so you can hold a spoken conversation with an AI without any server requests. The whole pipeline runs locally on your GPU via WebGPU, making it suitable for exploring on‑device conversational agents.
Can I compare multiple TTS or STT models at once?
Yes; TTSLab provides a side‑by‑side comparison mode where you can run two or more TTS or STT models on the same input text or audio. This lets you directly hear and measure differences in quality, latency, and style across models in a standardized environment.
Is TTSLab suitable for production or only for testing?
TTSLab is primarily designed as a testing and evaluation platform rather than a production deployment system. It is ideal for benchmarking, auditioning models, and exploring on‑device voice AI, but it does not provide managed hosting, scaling, or enterprise‑grade SLAs like cloud TTS/STT services.