Top AI Product

What VibeVoice actually is

Microsoft open-sourced a long-form, multi-speaker text-to-speech model — and it blew up to 51K stars. This isn’t a demo toy. VibeVoice synthesizes up to 90 minutes of continuous audio with as many as 4 distinct speakers in a single pass, in English and Chinese. The trick is a continuous speech tokenizer running at a 7.5 Hz frame rate: it keeps fidelity high while making long sequences cheap to generate. The base is Qwen2.5 1.5B.

The pitch writes itself. ElevenLabs-grade naturalness and emotion, except the weights are yours to self-host. No per-token cloud bill for every podcast episode, audiobook chapter, or agent reply.

The API angle

A newer VibeVoice-Realtime 0.5B (MIT license) streams text in and emits first audio in about 300ms, with a WebSocket API for real-time use — voice agents, live narration, dialogue products. Self-host it, feed streaming text, get 24kHz speech back.

One caveat: Microsoft pulled the original TTS code after it found misuse, so most people now run community forks of the weights.

You Might Also Like

• Microsoft Vibevoice Scores 24k Github Stars Doing What Elevenlabs Charges 99 Month for
• Ai Agent Book Bojie lis Open Source ai Agent Textbook Gains 1734 Github Stars in one day
• Hermes Agent by Nous Research Might be the Open Source ai Agent That Finally Remembers Everything
• Langchains gtm Agent Drove 250 More Conversions now the Framework Behind it is Open Source
• Onyx Hits 19 7k Github Stars the Open Source Answer to Gleans 7 2b Enterprise ai Play