CosyVoice2: FSQ tokens and chunk-aware streaming flow synthesis

What changed

Combined finite-scalar speech-token quantization, a pretrained text-language backbone and chunk-aware causal flow rendering to support streaming and non-streaming speech in one system.

Scientific scope

Combined finite-scalar speech-token quantization, a pretrained text-language backbone and chunk-aware causal flow rendering to support streaming and non-streaming speech in one system. Inherits CosyVoice, FSQ and flow matching. Distinct delta is the causal/chunk/token interface, not model-size release. Source human-parity and virtually-lossless streaming claims are empirical, not lossless coding or universal quality guarantees.

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 CosyVoice, FSQ and flow matching. Distinct delta is the causal/chunk/token interface, not model-size release. Source human-parity and virtually-lossless streaming claims are empirical, not lossless coding or universal quality guarantees.

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 2: Scalable Streaming Speech Synthesis with Large Language Models

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