LLM-based error correction improves zero-shot Whisper and fine-tuned WavLM child ASR transcriptions, but not fine-tuned Whisper, and conversational context as implemented degrades corrections.
First, larger LLMs consistently improve zero-shot ASR perfor- mance across different Whisper models, while smaller LLMs show limited benefits
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Large Language Models based ASR Error Correction for Child Conversations
LLM-based error correction improves zero-shot Whisper and fine-tuned WavLM child ASR transcriptions, but not fine-tuned Whisper, and conversational context as implemented degrades corrections.