Prompting Whisper with synthetic-error text and prompting an LLM with aligned hypotheses cuts Dutch child read speech WER from 9.4% to 5.1% and lifts reading mistake detection F1 from 0.39 to 0.73.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Child Speech Recognition and Reading Mistake Detection by Using Prompts
Prompting Whisper with synthetic-error text and prompting an LLM with aligned hypotheses cuts Dutch child read speech WER from 9.4% to 5.1% and lifts reading mistake detection F1 from 0.39 to 0.73.