# Eloquent Web

> Speak naturally, get polished text. 100% on-device AI dictation powered by Cactus Whistle streaming STT and Superwhisper S1-mini via WebAssembly and WebGPU.

## Links
- [App](https://h3manth.com/ai/eloquent/): Live Eloquent Web application
- [Source](https://github.com/hemanth/eloquent-web): GitHub repository
- [Author](https://h3manth.com): Hemanth HM
- [Whistle Model](https://huggingface.co/Cactus-Compute/whistle): Cactus Whistle streaming speech-to-text model
- [S1-mini Model](https://huggingface.co/onnx-community/s1-mini-ONNX): Superwhisper S1-mini inverse text normalizer

## Features
- Real-time on-device streaming speech-to-text (~18 MB WebAssembly via Needle 3)
- AI transcript polishing, disfluency removal, and self-correction resolution
- One-tap prose transformations: Polish, Key points, Formal, Short
- Offline-capable IndexedDB model caching with zero server/cloud audio uploads

## Tech stack
- Cactus Whistle (`Cactus-Compute/whistle` + `Cactus-Compute/needle3` WASM)
- Superwhisper S1-mini (`onnx-community/s1-mini-ONNX`) via Transformers.js 4.2.0 WebGPU
- Vanilla ES Modules, Web Workers, Web Audio API, and Material Design 3 CSS