---
title: Speaker Diarization for Returning Voices in Long Meetings
seo_title: Speaker Diarization for Returning Voices in Long Meetings
description: MOSS-Transcribe-Diarize 0.9B is a strong open-source candidate when a
  long meeting needs words, timestamps, and speaker labels together. Review the documented
  evidence and test the workflow with your own approved materials.
topic: Speech Transcription and Diarization
date: '2026-11-12'
lastmod: '2026-11-12'
author: Model Workflow Lab Editorial
author_type: Organization
faq:
- q: Which version scored 15.83%?
  a: The published AISHELL-4 figure is for MOSS-Transcribe-Diarize 0.9B.
- q: Can cpCER verify continuity over a particular gap?
  a: Review the relevant turns directly and record split and merge errors.
- q: Does a speaker tag identify a person?
  a: Map anonymous labels to participants using verified recording context.
category: library
copyright: Copyright 2026 ModelWorkflowLab.com; subject to stated permissions and
  exceptions.
---
# Speaker Diarization for Returning Voices in Long Meetings

MOSS-Transcribe-Diarize 0.9B is a strong open-source candidate when a long meeting needs words, timestamps, and speaker labels together. The published AISHELL-4 comparison reports **15.83% cpCER** for 0.9B, **24.99%** for VibeVoice ASR, and **27.43%** for Gemini 3 Pro. The separate MOSS-Transcribe-Diarize Pro row reports 14.02%. [OpenMOSS evaluation](https://github.com/OpenMOSS/MOSS-Transcribe-Diarize)

<figure>
<img src="/assets/images/diarization-returning-speakers-long-meetings.svg" alt="Mark early turns, then Find later returns, then Check label continuity" loading="lazy" width="1200" height="200" />
<figcaption>Proposed evaluation workflow.</figcaption>
</figure>

## Test a speaker who returns after a long gap

The public AISHELL-4 result uses meeting audio. We would extend that case into a continuity audit with a participant who speaks early, stays silent for several agenda items, and then returns. Map anonymous labels once for the entire file. Remapping each excerpt independently could hide a label split. [Technical report](https://arxiv.org/abs/2601.01554)

| Observation | Review decision |
|---|---|
| Same person receives a new label | Count a split |
| Two people share a label | Count a merge |
| Correct label with incorrect words | Log a recognition error |
| Two voices speak simultaneously | Apply the overlap policy |

The official dialogue examples show time ranges and anonymous speaker tags. Use the recording to establish identity and inspect turns around the gap. cpCER evaluates speaker-grouped text after label matching. A timeline-only application should score activity boundaries separately. [Official demos](https://www.moss.cn/models/moss-transcribe-diarize) · [Checkpoint](https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize)

The release also reports first place in Task 1 of the second MLC-SLM Challenge at INTERSPEECH 2026. Keep the competition claim tied to its task. See the [challenge report](https://arxiv.org/abs/2609.27514) and the [overlap review](/speaker-diarization-overlapping-dialogue/).

## English SEO keywords

returning speakers, speaker label continuity, AISHELL-4 cpCER.