Peer-reviewed research
Generative Speech Recognition Error Correction with Large Language Models and Task-Activating Prompting
Chao-Han Huck Yang et al., IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) 2023, 2023
Read the originalRead on October 17, 2026
- What it is
- The authors use large language models to correct speech recognition output, by rescoring candidate transcripts and by generating corrections, and test on two out-of-domain tasks, ATIS and WSJ.
- How much weight it carries
- Accepted to the ASRU 2023 workshop. The abstract reports the direction of the results; I did not read the full error-rate tables. It tests 2023 recognizers on two datasets, so newer systems may behave differently.
Facts taken from this source
- Yang and colleagues study whether large language models can act as post-processors for speech recognition, performing rescoring and error correction on the output of a recognizer.
“We explore the ability of large language models (LLMs) to act as speech recognition post-processors that perform rescoring and error correction.” (Abstract)
- Amazon researchers report that combining LLM prompting with fine-tuning brought speech recognition error rates below the N-best oracle level, a benchmark for choosing among the recognizer's candidate transcripts.
“By combining prompting techniques with fine-tuning we achieve error rates below the N-best oracle level” (Abstract)