Reverb-TTS open-sourced: zero-shot voice cloning in-browser on WebGPU, 17x realtime on M4 Pro
FinancialAd1961 · reddit · 2026-11-20
Software Mansion open-sourced Reverb-TTS, a zero-shot voice cloning model built for fully local inference. It runs in the browser via WebGPU on their ruNNtime engine (built on TypeGPU), hitting 17x realtime on a MacBook M4 Pro with nothing sent to a server. The demo runs three models on GPU simultaneously — LFM-3 for frame description, Whisper-small for reference transcription, Reverb for cloning — with video frames fed straight to GPU. Model card, benchmarks, and a HF Space demo are available.
More from Infra
- Open weights jump from 7.3% to 30.9% of Databricks-hosted tokens in four months — AIFlow_ML · 2026-11-20
- 7 web search APIs with MCP servers for Claude and Cursor, benchmarked — JeremyCMorgan · 2026-11-20
- Baseten ships deferred tool loading: agents run up to 62% cheaper and 26% faster — baseten · 2026-11-20
- AI racks too heavy for old datacenters: GB300 hits 1,590kg, Vera Rubin ~1,814kg — terryyuezhuo · 2026-11-20
- Point-by-point rebuttal: Hunterbrook's AMD report ignores publicly downloadable BIOS firmware — firstadopter · 2026-11-20
- Touchmark raises $3M seed to build a commodities-style exchange for AI inference contracts — ycombinator · 2026-11-20