# Group Relative Policy Optimization for Speech Recognition

arXiv: https://arxiv.org/abs/2509.01939 · HTML: https://mlsanity.com/paper/2509.01939/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 2 Sep 2025.

## Built on by (most cited first, 8 of 8)

- [TLive-Omni: An Omni-Modal Understanding Model for E-Commerce Live Streaming](https://mlsanity.com/paper/2608.20958/) (2026)
- [LEAF: Growing Trees Without Branching for Speech-Aware Large Language Model Post-Training](https://mlsanity.com/paper/2606.07610/) (2026)
- [Reinforcement Learning for Data-Efficient Code-Switched ASR](https://mlsanity.com/paper/2607.02757/) (2026)
- [TokAN: Accent Normalization Using Self-Supervised Speech Tokens](https://mlsanity.com/paper/2607.03928/) (2026)
- [When Synthetic Speech Is All You Have: Better Call GRPO](https://mlsanity.com/paper/2607.08409/) (2026)
- [Towards Quantifying Benchmark Optimization in ASR Models](https://mlsanity.com/paper/2608.19936/) (2026)
- [Look Less, Hear Better: Jointly Rewarded GRPO for Streaming ASR](https://mlsanity.com/paper/2609.18333/) (2026)
- [Transcribe, Translate, and Optimize: Joint Reward Learning for Speech Translation](https://mlsanity.com/paper/2609.26536/) (2026)

## Abstract (arXiv)

Speech Recognition has seen a dramatic shift towards adopting Large Language Models (LLMs). This shift is partly driven by good scalability properties demonstrated by LLMs, ability to leverage large amounts of labelled, unlabelled speech and text data, streaming capabilities with auto-regressive framework and multi-tasking with instruction following characteristics of LLMs. However, simple next-token prediction objective, typically employed with LLMs, have certain limitations in performance and challenges with hallucinations. In this paper, we propose application of Group Relative Policy Optimization (GRPO) to enable reinforcement learning from human feedback for automatic speech recognition (ASR). We design simple rule based reward functions to guide the policy updates. We demonstrate significant improvements in word error rate (upto 18.4% relative), reduction in hallucinations, increased robustness on out-of-domain datasets and effectiveness in domain adaptation.