Voice of Reason: verifiable-reward training for spoken mathematical reasoning

What changed

Adapted a speech-native model with synthetic spoken question/answer supervision and verifiable-reward reinforcement learning, studying reasoning-token and streaming tradeoffs.

Scientific scope

Adapted a speech-native model with synthetic spoken question/answer supervision and verifiable-reward reinforcement learning, studying reasoning-token and streaming tradeoffs. Inherits GLM4Voice, reinforcement learning with verifiable rewards and streaming reasoning. Solving spoken benchmark exercises is not a new mathematical discovery/proof or general reasoning guarantee; speech synthesis/transcription errors and benchmark scope remain.

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 GLM4Voice, reinforcement learning with verifiable rewards and streaming reasoning. Solving spoken benchmark exercises is not a new mathematical discovery/proof or general reasoning guarantee; speech synthesis/transcription errors and benchmark scope remain.

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.

• Voice of Reason: Reinforcement Learning for Spoken Math

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