Skip to content
Preprint

SpeechGym: An Audio-Native Gym for Training Voice Agents via Reinforcement Learning

Aug 2026 · 1 citation · ⚡ 1 influential · 39 references
Computer Science

TL;DR

SpeechGym is presented, an audio-native agentic environment in which two omni-modal models converse in native audio, with no external ASR or TTS and no API boundary, over the unmodified tasks, tools and success check of an established text agentic benchmark, so that the interaction modality is the only variable and the loop stays local and trainable end to end.

Abstract

Voice agents must call tools and hold multi-turn dialogue entirely through speech, yet the dominant paradigm trains them in text. Existing frameworks either cascade TTS and ASR around a proprietary voice API, where gradients cannot flow and per-call cost makes on-policy reinforcement learning prohibitive, or stay in text: they measure voice agents but cannot improve them. We present SpeechGym, an audio-native agentic environment in which two omni-modal models converse in native audio, with no external ASR or TTS and no API boundary, over the unmodified tasks, tools and success check of an established text agentic benchmark, so that the interaction modality is the only variable and the loop stays local and trainable end to end. Audio agentic capability does not follow from audio understanding. The failures speech introduces are perceptual rather than reasoning deficits: the agent picks the right tool and the right argument slot but fills it with a value misheard from the waveform, and that single error cascades into a failed call, a retry of the same call, and a wasted step budget. A second failure is behavioural: under an insistent caller the agent performs an unauthorised write and ends the episode believing it helped. Both are trainable, because the environment labels them for free: a call with a misheard argument fails against the database while a correct one succeeds. The obstacle is sparsity, not signal. Outcome-only GRPO is gradient-starved here, since almost every rollout group fails identically, while a per-turn process reward crediting each successful tool call restores variance to nearly every group. Trained this way, the agent transfers with no further tuning to an independently implemented voice benchmark, more than doubling task success and carrying an open-weights model from last place to second on that leaderboard, while using fewer turns and tokens than before training.

View source

Similar papers

#artificial intelligence Preprint Sep 2026

Qwen-Audio-3.1-Realtime: Towards Reliable Agentic Voice Interaction

Real-time voice assistants must reason over evolving requests, execute actions, and follow conversational rules. Qwen-Audio-3.1-Realtime brings these requirements together through Think, Act, and Speak and Coordinate. Think combines Core-Cocktail supervised fine-tuning with Multimodality and Multi-Teacher On-Policy Dis...

Lu-Jia Bao, Qian Chen, Luyao Cheng et al. · 1 citation
Preprint Sep 2026

Listen, Critique, and Refine: RL-Based Self-Refinement for Instruction-Following Speech Synthesis

This work training a LALM with reinforcement learning to reason over its own speech output generates a draft speech as a form of audio-token reasoning, critiques its own generation by reflecting on the acoustic realization in text, and produces a refined version conditioned on both the first-pass speech and the critiqu...

Chee-En Yu, Yi-Cheng Lin, Sung-Feng Huang et al. · 1 citation

Voice of Reason: Reinforcement Learning for Spoken Math

This work first adapts the model to the domain using supervised fine-tuning on synthesized spoken question-answering data and shows that RL improves the accuracy on GSM8K beyond levels previously achieved for speech models only with supplementary reasoning traces, and establishes a new state-of-the-art for mathematical...

Timothée Weisselberger, Edouard Graves, Alexandre Défossez · 0 citations
Preprint Aug 2026

Motion-Omni: End-to-End Joint Speech and Full-Body Motion for Spoken Dialogue

Motion-Omni is presented, an end-to-end framework in which a spoken dialogue model natively outputs explicit facial expression together with hand, upper-body and lower-body motion, generated directly from the hidden states that produce the speech.

Chengqian Ma, Wei Tao, Hao-Yu Zhang et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.