CosyVoice3: multitask speech tokenization and differentiable-reward post-training

What changed

Extended supervised speech token learning across semantic, prosodic and speaker tasks and introduced differentiable-reward post-training alongside a broader training corpus.

Scientific scope

Extended supervised speech token learning across semantic, prosodic and speaker tasks and introduced differentiable-reward post-training alongside a broader training corpus. Inherits CosyVoice2 and multitask/reward learning. Acquired delta is specified tokenizer/post-training mechanism, not generic scaling of hours/parameters/languages. Reported multilingual quality is not universal robustness or independently validated human parity.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits CosyVoice2 and multitask/reward learning. Acquired delta is specified tokenizer/post-training mechanism, not generic scaling of hours/parameters/languages. Reported multilingual quality is not universal robustness or independently validated human parity.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• CosyVoice 3: Towards In-the-wild Speech Generation via Scaling-up and Post-training

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction