analiza aiqbench
This research demonstrates that Group Relative Policy Optimization (GRPO) significantly outperforms supervised fine-tuning when adapting large language models to audio tasks using only synthetic speech data. By leveraging reinforcement learning, the method reduces word error rates by up to 45% in regulated domains like banking where real-world recordings are legally restricted. The findings suggest that behavior-based optimization is more effective than representation learning for bridging the acoustic gap between synthetic and real audio without requiring additional model layers.
When Synthetic Speech Is All You Have: Better Call GRPO
arXiv:2607.08409v2 Announce Type: replace Abstract: LLM-based ASR adapted to regulated domains such as banking is bottlenecked by privacy: real speech is costly and legally constrained to collect, making synthetic text-to-speech (TTS) an attractive substitute. Yet synthetic speech stays acoustically mismatched with real recordings, and work on this gap has stayed within supervised fine-tuning (SFT). We instead turn to reinforcement learning, and show that Group Relative Policy Optimization (GRPO) extracts far more from the same synthetic speech than SFT. Synthetic-only adaptation of the model with GRPO, a critic-free method rewarding low-WER hypotheses, reduces WER by 40% relative to SFT (36.71%$\to$22.09%), and an SFT-then-GRPO combination pushes this further to 45%. We trace the gain to behavior rather than representation: GRPO reduces insertion errors by improving stopping calibration and speech-to-text alignment by better anchoring attention to audio, leaving early-layer representations intact. When synthetic speech is the main resource, reinforcement learning should be preferred over supervised fine-tuning.
Abstract:LLM-based ASR adapted to regulated domains such as banking is bottlenecked by privacy: real speech is costly and legally constrained to collect, making synthetic text-to-speech (TTS) an attractive substitute. Yet synthetic speech stays acoustically mismatched with real recordings, and work on this gap has stayed within supervised fine-tuning (SFT). We instead turn to reinforcement learning, and show that Group Relative Policy Optimization (GRPO) extracts far more from the same synthetic speech than SFT. Synthetic-only adaptation of the model with GRPO, a critic-free method rewarding low-WER hypotheses, reduces WER by 40% relative to SFT (36.71%$\to$22.09%), and an SFT-then-GRPO combination pushes this further to 45%. We trace the gain to behavior rather than representation: GRPO reduces insertion errors by improving stopping calibration and speech-to-text alignment by better anchoring attention to audio, leaving early-layer representations intact. When synthetic speech is the main resource, reinforcement learning should be preferred over supervised fine-tuning.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2607.08409 [cs.CL]
(or arXiv:2607.08409v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2607.08409 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 244 cuvinte
Subiecte conexe
SyRA-CS: Synthetic-to-Real Adaptation for Code-Switched ASR with Meta-Learning
arXiv · 24 oct. 2026
LibriConvo: Simulating Conversations from Read Literature for ASR and Diarization
arXiv · 11 iun. 2026
Building and Evaluating a Synthetic Bengali Speech Resource for Telecom Customer Care
arXiv · 24 aug. 2026
ArFake: A Robust Framework for Multi-Dialect Arabic Speech Spoofing Detection Benchmark
arXiv · 17 iun. 2026