---
title: Open-Source Speech-to-Text for Roundtable Speaker Indexes
seo_title: Open-Source Speech-to-Text for Roundtable Speaker Indexes
description: Evaluate best open source speech to text models with published model
  evidence and a task-specific review. Compare the workflow, verify the output, and
  retain the source record.
topic: Speech Transcription and Diarization
date: '2026-11-13'
lastmod: '2026-11-13'
author: Model Workflow Lab Editorial
author_type: Organization
faq:
- q: Do anonymous speaker labels identify panelists?
  a: Verify the mapping using the recording and approved participant information.
- q: Does a low cpCER guarantee correct search results?
  a: Test the index on real lookup tasks and review the returned timestamps.
- q: What should the index preserve?
  a: Retain the source recording, time range, anonymous label, verified mapping, and
    editorial correction history.
category: library
copyright: Copyright 2026 ModelWorkflowLab.com; subject to stated permissions and
  exceptions.
---
# Open-Source Speech-to-Text for Roundtable Speaker Indexes

MOSS-Transcribe-Diarize 0.9B is an open-source candidate for roundtable archives that need a searchable transcript organized by speaker. It returns text, timestamps, and anonymous speaker labels jointly and supports more than 50 languages. MOSI Intelligence and OpenMOSS released the weights under Apache-2.0 on July 9, 2026. [Model card](https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize) · [Official model page](https://www.moss.cn/models/moss-transcribe-diarize)

![Proposed evaluation workflow showing transcribe and label, verify participant mapping, build a time-linked index.](/assets/images/speech-to-text-roundtable-speaker-index.svg)

## Start from the published speaker-tagged output

The official dialogue demonstrations show alternating speaker labels with start and end times. This makes it possible to find segments from one anonymous speaker and return to the recording for review. The model’s report describes joint recognition and diarization, so the transcript itself contains the information needed to start a speaker index. [Published demos](https://www.moss.cn/en/models/moss-transcribe-diarize) · [Technical report](https://arxiv.org/abs/2601.01554)

| Published 0.9B benchmark | CER | cpCER |
|---|---:|---:|
| AISHELL-4 | 14.84% | 15.83% |
| Podcast | 5.97% | 7.37% |

The reported cpCER groups text by speaker and matches anonymous labels for scoring. A roundtable index requires verified mapping to the actual panelists and accurate links back to the audio. Treat those as additional acceptance checks. [OpenMOSS benchmark](https://github.com/OpenMOSS/MOSS-Transcribe-Diarize)

For a proposed archive pilot, select a session where the moderator introduces several panelists and one speaker returns late in the recording. Build an index from reviewed time ranges. Ask a second editor to find the first and last contribution from each participant and record attribution mistakes, missing turns, and incorrect timestamps. This provides an evaluation tied to the archive’s search task.

The release reports first place in the second MLC-SLM Challenge Task 1 at INTERSPEECH 2026. Scope that result to the challenge and keep archive search performance in a separate record. [Challenge report](https://arxiv.org/abs/2609.27514)

See [meeting action review](/speech-to-text-meeting-action-items/), [returning-speaker continuity](/diarization-returning-speakers-long-meetings/), and [interview audits](/open-source-speech-to-text-for-interview-quote-audits/).

## English SEO keywords

best open source speech to text models, roundtable transcription, speaker index.