FrankenTTS: Pure Rust Port of Qwen3-TTS Runs Zero-Shot Cloning in Browser
doodlestein · x · 2026-08-10
A developer introduced FrankenTTS, a pure Rust port of the Qwen3-TTS-12Hz-0.6B voice cloning model. The project completely abandons Python and PyTorch dependencies, requiring no server or GPU.
- Local Execution: The model is compiled to WebAssembly and runs entirely locally in the user's browser. No text or audio leaves the tab, ensuring strict privacy.
- Performance: On a native Mac CLI, synthesis runs at 1.4–1.6× real time; in the browser, it currently achieves 0.31–0.43× real time, with further optimizations underway.
- Core Features: Supports zero-shot voice cloning. Users can record a short enrollment script directly on the page, and the system extracts a denoised 4KB voice vector to generate custom speech.
More from coding & agent
- Open Source Tool Clay: Multiplayer Workspace for Claude Code and Codex — tom_doerr · 2026-08-10
- Stanford's CS329A Self-Improving AI Agents Course Released on YouTube — dhruv2038 · 2026-08-10
- Don't Trust High Resolution Rates: Contain AI Agents in Docker Sandboxes — BarracudaMean9308 · 2026-08-10
- AI Agent Earns $14 Autonomously, Developer Calls It a Personal AGI Moment — koltregaskes · 2026-08-10
- Disabling Smart Memory in ComfyUI Boosts Minimax H3 Generation Speed by 65-70% on RTX 3090 — Life_is_important · 2026-08-10
- New Agent Auditing Engine Reveals 10x Token Cost Gap Between Frameworks — alex_verem · 2026-08-10