Fact · Speech recognition
A 2025 study found that a substantial amount of the LibriSpeech and Common Voice evaluation sets appear in public large language model pretraining corpora.
Checked against its source on October 17, 2026
In the source's own words
“this work finds that a substantial amount of the LibriSpeech and Common Voice evaluation sets appear in public LLM pretraining corpora”
Evaluation of LLMs in Speech is Often Flawed: Test Set Contamination in Large Language Models for Speech Recognition, Yuan Tseng et al., arXiv (Audio and Speech Processing), 2025, Abstract.
Open the original- What it measures
- The authors checked the text of public LLM pretraining corpora against the evaluation sentences of the two corpora most often quoted in speech papers.
- What it does not show
- The finding is about text: sentences from the test sets appear in pretraining text. It does not show that any specific commercial model was trained on the test audio.