sts-web
A full-duplex speech-to-speech assistant running entirely client-side, still early and rough.
Browser-native speech-to-speech, 100% client-side via Rust/WASM + WebGPU. Runs a pruned 24-layer, Q4_K-quantized PersonaPlex-7B (QLoRA-recovered from NVIDIA’s 32-layer original) through a full-duplex pipeline: mic to Mimi encoder to temporal/depth transformers to Mimi decoder. Walkie-talkie mode works with a handful of voice presets; true full-duplex streaming is not supported yet, and audio quality is currently poor.
Includes
model
nvidia/personaplex-7b-v1, by NVIDIA
Pruned, QLoRA-recovered, Q4_K quantization for WebGPU by Idle Intelligence.
NVIDIA Open Model License
modelnvidia/personaplex-7b-v1, by NVIDIA
Q4_K quantization for WebGPU by Idle Intelligence.
NVIDIA Open Model License
- Runs
- browser, WASM, WebGPU
- Demo (GitHub Pages)
- https://idle-intelligence.github.io/sts-web/web/
- Source
- idle-intelligence/sts-web
- HuggingFace
- idle-intelligence/personaplex-24L-q4_k-webgpu
- Status
- experiment
Learn more
Used on: sts