VibeVoice 7B

VibeVoice is Microsoft's open-source family of frontier voice models, released August 2025, built for the problem most TTS systems fumble: long-form, multi-speaker speech. The 7B TTS variant synthesizes up to 90 minutes of continuous audio with as many as four distinct speakers, targeting scripted podcasts, training modules, and dialogue-heavy explainers. It pairs a Qwen2.5-based language model for dialogue structure with continuous speech tokenizers running at an ultra-low 7.5 Hz frame rate and a next-token diffusion head for acoustic detail.

The honest picture, and it matters: on September 5, 2025 Microsoft removed the VibeVoice-TTS code from its repository after discovering uses inconsistent with the project's stated research intent — the weights still circulate via community mirrors, but the official distribution channel is gone and future updates are uncertain. Voice-cloning capability this accessible carries real misuse risk, which is worth weighing before building on it. A smaller 1.5B variant and a 0.5B real-time streaming version round out the family.