AI·LLM-based ASR in regulated domains like banking is limited by privacy and real-speech collection costs; synthetic TTS data is a cost-effective substitute, but acoustic mismatch hinders supervised fine-tuning (SFT). Group Relative Policy Optimization (GRPO) applied solely to synthetic speech reduces WER by 40% relative to SFT (36.71% to 22.09%) and by 45% in the SFT-then-GRPO sequence, by improving behavioral calibration and audio attention rather than representations.
Search
What are you looking for?
2 results for “reinforcement-learning”
AI·The paper identifies a failure mode in critic-free RL methods for LLMs where implausible tokens receive uniform credit, and proposes tail-aware credit calibration to improve reinforcement learning.