LLM agent improves terminology accuracy in long speech transcripts
agentic-ger: terminology recovery in long-form speech using global context
Computation and LanguageArtificial Intelligence
Summary
Transcribing long recordings accurately is hard, especially for specialized words. The authors created an AI helper that looks at the entire transcript to spot and fix unusual or unclear terms. It even listens again to certain parts to confirm better guesses. This approach helped improve transcription accuracy in both Chinese and English recordings compared to existing systems.
What this means in practice
- •For speech recognition engineers: Improve domain-specific word accuracy in long audio transcripts by integrating an LLM-based terminology correction agent.
- •For multilingual transcription service providers: Enhance Chinese and English transcription quality for specialized vocabulary through global context informed corrections.$Commercial implications: Enables offering higher-accuracy transcription services for domain-specific content, attracting clients needing specialized terminology handling.
Authors
Yanqiao Zhu, Wupeng Wang, Zhifu Gao, Xiangang Li, Xie Chen
Abstract
Recent advances in speech language models have improved automatic speech recognition (ASR) for long-form audio. However, accurately and consistently transcribing domain-specific terminology remains challenging. Motivated by the world knowledge and contextual capability of large language models (LLMs), we propose Agentic-GER, an LLM-based agent for terminology correction in long-form speech. The agent uses global context from the full transcript to identify suspicious terms and resolve ambiguous hypotheses. It selectively re-transcribes the source speech to check candidate corrections, and uses accepted edits to guide subsequent decisions. Experiments with four LLMs and two ASR systems on GigaSpeechBench show consistent terminology improvements in both Chinese and English, with and without thinking. On Chinese speech, Agentic-GER achieves up to a 36.8% relative reduction in biased character error rate (B-CER) over the Whisper baseline.