# Transcribe a recorded meeting or call and say who said what

> Transcribe a recorded meeting or call and say who said what: 6 companies fully do it, 10 with a limit, 45 carry the tag. Every listing in Speech-to-text with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/speech-to-text.meeting-with-speakers/
- Markdown: https://www.anchorterminal.com/niches/speech-to-text.meeting-with-speakers/index.md (~2,000 tokens)
- Slim: https://www.anchorterminal.com/niches/speech-to-text.meeting-with-speakers/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/speech-to-text.meeting-with-speakers/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 6 | 10 | 45 | 20 | 9 |

## What counts as done

A transcript with a stable speaker label on every segment, labels consistent across the whole file including any chunks, in one job.

Measured, for our own tests: speaker count within one of the true count.

How agents ask for it: transcribe this meeting and label the speakers; who said what in this call; give me notes with each action item attributed

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Amazon Transcribe (Amazon Web Services, Inc.)**: ShowSpeakerLabels with MaxSpeakerLabels 2 to 30 on batch jobs; files to 28,800 seconds and 2 GB, so a 3-hour meeting runs as one job without chunking.
- **AssemblyAI Speech-to-Text (Universal) (AssemblyAI, Inc.)**: speaker_labels with speakers_expected or min and max caps on files to 10 hours and 5 GB in one job; labels are generic A, B (acc-03) and acc-30 lost text on the legacy tier.
- **Azure AI Speech speech-to-text (Microsoft Corporation)**: Fast transcription takes diarization maxSpeakers 2 to 35 on files under 5 hours and 500 MB; batch transcription has diarizationEnabled and 1 GB files.
- **ElevenLabs Scribe Speech to Text API (Eleven Labs Inc.)**: diarize with num_speakers up to 32 on files to 5 GB and 15 hours in one request; use_speaker_library matches registered speakers across jobs.
- **Rev AI Speech-to-Text API (Rev.com, Inc.)**: Diarisation is on unless skip_diarization is set, on media to 17 hours and 2 GB in one job; no speaker cap is documented.
- **Speechmatics Speech-to-Text (Cantab Research Ltd)**: diarization speaker with speaker_diarization_config.max_speakers (unrestricted if omitted) and sensitivity; batch files to 1 GB inline or larger via fetch_data URL. acc-01 ranked it last on postcodes, not speakers.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Deepgram Speech-to-Text (Nova-3, Flux) (Deepgram, Inc.)**: diarize_model v2 labels every word on files to 2 GB in one request, but no speaker cap is documented and acc-29 saw accuracy drop past four speakers.
- **Gladia Speech-to-Text API + MCP (Gladia SAS)**: diarization with number_of_speakers, min and max hints (not enforced) on pre-recorded files to 135 minutes and 1,000 MB; a 3-hour meeting must be split, which restarts speaker labels.
- **Google Cloud Speech-to-Text (Google LLC)**: diarization_config with min and max speaker counts; Chirp 3 diarises only in BatchRecognize from Cloud Storage, files to 8 hours. acc-11 says live diarisation stops at about three speakers.
- **IBM Watson Speech to Text (International Business Machines Corporation)**: speaker_labels works on HTTP and WebSocket requests to 1 GB async, but is optimised for two speakers with a six-speaker maximum and undefined behaviour beyond.
- **Inworld STT (Theai, Inc. dba Inworld AI)**: enableSpeakerDiarization is experimental, applies to the streaming endpoint, and labels per stream; async jobs to 512 MB say diarisation works as synchronously, with no speaker cap.
- **MiniMax Speech-to-Text (Nanonoble Pte. Ltd.)**: verbose_json returns n_speakers and speaker on each segment, but requests stop at 500 seconds, so any meeting must be split and speaker labels restart per chunk.
- **Mistral Voxtral Transcribe (Mistral AI (RCS Paris 952 418 325))**: diarize with timestamp_granularities segment gives speaker_id per segment on batch; the 3-hour limit is a vendor claim and the card says one speaker is kept on overlap. acc-09 found it good for podcasts.
- **OpenAI Speech to Text (OpenAI OpCo, LLC)**: Only the deprecated gpt-4o-transcribe-diarize returns diarized_json, within 25 MB per file, so long meetings are chunked and labels flip per chunk (acc-20); it shuts on 2027-02-26.
- **Smallest.ai Pulse STT (Smallest, Inc.)**: diarize adds per-word and per-utterance speaker labels on pre-recorded files, but no speaker cap is stated and Pulse Pro caps audio at 250 MB.
- **Soniox Speech-to-Text (Soniox Inc.)**: enable_speaker_diarization labels every token on async files to 300 minutes, so a 3-hour file is split; no speaker cap is stated. acc-11 says Soniox does live speakers well.

- not this job: Cartesia Ink (Cartesia AI, Inc.): No diarisation parameter on either endpoint; batch POST /stt accepts ink-whisper only.
- not this job: Cohere Transcribe 03-2026 (Cohere Inc.): The docs state the model has neither timestamps nor speaker diarisation.
- not this job: Giggy Speech to Text (Giggy): No speaker labels; the docs list speaker labels among the functions the API does not return.
- not this job: Groq Speech-to-Text (Groq LLC): No diarisation parameter on the OpenAI-compatible endpoints.
- not this job: Ming-3.0-asr-tiny (inclusionAI, Ant Group (named in the site footer of inclusion-ai.org; the LICENSE copyright line reads inclusionAI)): Card states output has no speaker diarisation.
- not this job: Ming-3.0-audio-flash (inclusionAI, Ant Group (named in the inclusion-ai.org footer; no registered company named)): No diarisation or speaker output on the card.
- not this job: Ming-3.0-audio-tiny (inclusionAI (Ant Group)): No diarisation on the card.
- not this job: Nemotron 3.5 ASR Streaming 0.6B (NVIDIA Corporation): No diarisation; plain text out.
- not this job: Parakeet TDT 0.6B v3 (NVIDIA Corporation): No diarisation; acc-12 paired it with a separate diarisation model.
- not this job: Qwen3-ASR-0.6B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check)): No diarisation; acc-08 chained pyannote separately.
- not this job: Qwen3-ASR-1.7B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check)): No diarisation; acc-08 chained pyannote separately.
- not this job: Underdog Bark 3.0 (Conway Research): Plain text only, no speaker labels.
- not this job: Voxtral Mini 4B Realtime 2602 (Mistral AI (RCS Paris 952 418 325)): No diarisation or segments for the weights.

3 listings in the category were added after this reading and aren't marked yet.

## What stops agents finishing

- diarisation restarts per chunk when the file must be split (OpenAI 25 MB cap, acc-20)
- no diarisation at all (Groq, Giggy, Cohere Transcribe, Parakeet, Nemotron, Voxtral Realtime weights)
- speaker labels returned as A, B, C with no way to name them without a human step
- accuracy collapses past 4 speakers (acc-29)

## Around it

Before: meetings.record, storage.fetch-file. After: llm.summarise, document-extraction.action-items, notes.store.

## Where this comes from

Weight 20 of 100 in Speech-to-text: 10 of 32 accounts (acc-03, acc-06, acc-10, acc-11, acc-13, acc-20, acc-26, acc-29, acc-32, acc-09); the most common failure reported was attribution, not words

As data: https://www.anchorterminal.com/api/v1/use-cases/speech-to-text.meeting-with-speakers.json, or the MCP tool get_use_case with id speech-to-text.meeting-with-speakers. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
