audioshake/multispeaker2

AudioShake Multi-Voice demo

Static multitrack demo of Multi-Voice 2.0: one clean audio stream per speaker, clickable speaker-turn segments, and the two public per-frame confidence scores (assignment, separation) at 20 ms, drawn as a pair of rules under each turn.

Built with Vite + React + shadcn/ui; all data is pre-computed API-response JSON under data/. Source lives in the internal mv2-demo project.

Read more

• —Announcement — Multi-Speaker 2.0, keep every voice in the conversation
• —Technical write-up — how we evaluated it and where it stops working
• —Benchmarks — benchmarks and the full scoring config
• —Product page — Multi-Speaker separation, with the interactive demo