Preprint
End-to-End Joint ASR and Speaker Role Diarization with Child-Adult Interactions
Anfeng Xu et al., arXiv, 2026
Read the originalRead on October 16, 2026
- What it is
- A 2026 preprint on transcribing conversations between children and adults, including children assessed for autism with the ADOS, using Whisper-based models. It reports error rates by child age and by ADOS severity score.
- How much weight it carries
- Preprint, not yet peer-reviewed. The error rates are multi-talker transcripts of structured clinical and play sessions, not single-speaker dictation, so they indicate direction rather than the accuracy an adult would see.
Facts taken from this source
- In a 2026 preprint on child-adult recordings, children with high ADOS severity scores had higher multi-talker word error rates than children with low scores, within each age group.
“Across both age groups, children with high CSS consistently exhibit higher mtWER, whereas mtWER remains comparable within each age group at the same severity level.” (Experimental setup, Table V discussion)