AI
Who Said What, and Will It Be Remembered? Evaluating Persistent Speaker Attribution Across Meetings
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 08:27:25 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 17:18:24 UTC.
Briefings
This study evaluates persistent speaker attribution using the SI-cpWER metric across five commercial diarize-then-identify cascades, two academic baselines, and ThyVoice on the full CHiME-8 NOTSOFAR evaluation set. ThyVoice achieved the lowest mean SI-cpWER at 47.13, significantly outperforming all evaluated commercial systems which scored between 54.75 and 56.75. The research highlights that existing metrics fail to measure identity retention across meetings, necessitating direct evaluation of persistent attribution in systems that reuse conversations.
Key points
- ThyVoice achieved the lowest mean SI-cpWER at 47.13, significantly outperforming all evaluated commercial systems which scored between 54.75 and 56.75.
- Existing meeting-transcription metrics ignore speakers or remap anonymous speakers independently, preventing measurement of identity retention across meetings.
Context and limitations
Evaluation is limited to the full CHiME-8 NOTSOFAR evaluation set in clean and noiseaugmented forms, plus CHiME-6. Results are based on a single snapshot from September 2026 and do not account for future updates to commercial systems or changes in speaker identity over time.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 08:27:25 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidzrap5kd4clppfh4nuczlql2bn2vfvgq4x5zybyb6bp4lleighdy