# Produce captions or a word-aligned transcript

> Produce captions or a word-aligned transcript: 5 companies fully do it, 15 with a limit, 42 carry the tag. Every listing in Speech-to-text with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/speech-to-text.subtitles-timestamps/
- Markdown: https://www.anchorterminal.com/niches/speech-to-text.subtitles-timestamps/index.md (~1,800 tokens)
- Slim: https://www.anchorterminal.com/niches/speech-to-text.subtitles-timestamps/index.min.md (~430 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/speech-to-text.subtitles-timestamps/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 |
| --- | --- | --- | --- | --- |
| 5 | 15 | 42 | 8 | 3 |

## What counts as done

SRT or VTT, or JSON with a start and end time on every word, in one call.

Measured, for our own tests: no cue drifts more than 500 ms from the audio over an hour.

How agents ask for it: make subtitles for this video; give me word timestamps so I can cut clips; SRT for this recording

## Does the job

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

- **Amazon Transcribe (Amazon Web Services, Inc.)**: Subtitles Formats vtt and srt on StartTranscriptionJob, word timestamps in the JSON output, and files to 28,800 seconds in one job.
- **AssemblyAI Speech-to-Text (Universal) (AssemblyAI, Inc.)**: Word start and end on every word, GET /srt and /vtt with chars_per_caption, speaker_labels on the same job, files to 10 hours.
- **ElevenLabs Scribe Speech to Text API (Eleven Labs Inc.)**: timestamps_granularity word or character on every word, diarize on the same call, and additional_formats exports other formats; files to 15 hours.
- **Rev AI Speech-to-Text API (Rev.com, Inc.)**: GET /jobs/{id}/captions returns SRT or VTT, and the transcript carries per-word timestamps, on media to 17 hours.
- **Speechmatics Speech-to-Text (Cantab Research Ltd)**: format=srt on the transcript endpoint with srt_overrides, and start and end times on every word in the JSON.

## 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.

- **Azure AI Speech speech-to-text (Microsoft Corporation)**: Fast transcription returns words with offsets and batch has wordLevelTimestampsEnabled, but no srt or vtt output is documented on the API.
- **Cartesia Ink (Cartesia AI, Inc.)**: timestamp_granularities[] word returns start and end per word on ink-whisper batch; no srt or vtt output and no speaker labels.
- **Deepgram Speech-to-Text (Nova-3, Flux) (Deepgram, Inc.)**: Word start and end on every word and utterances, but no srt or vtt response format on the API; captions need a client library.
- **Gladia Speech-to-Text API + MCP (Gladia SAS)**: subtitles_config formats include srt with minimum and maximum duration, and words carry timestamps, but files over 135 minutes must be split.
- **Google Cloud Speech-to-Text (Google LLC)**: enable_word_time_offsets gives word times on Recognize and BatchRecognize, but Chirp 3 lists word timestamps as unsupported with degradation, no srt or vtt, and acc-06 found timestamps unreliable.
- **Groq Speech-to-Text (Groq LLC)**: verbose_json with timestamp_granularities word gives word times, but srt and vtt formats are not supported and files over the size cap must be split.
- **IBM Watson Speech to Text (International Business Machines Corporation)**: timestamps=true returns start and end per word (forced on with speaker labels), but no srt or vtt output.
- **Inworld STT (Theai, Inc. dba Inworld AI)**: includeWordTimestamps returns startTimeMs and endTimeMs per word, but no srt or vtt.
- **MiniMax Speech-to-Text (Nanonoble Pte. Ltd.)**: srt and vtt response formats with timestamp_level word, but only for files to 500 seconds.
- **Mistral Voxtral Transcribe (Mistral AI (RCS Paris 952 418 325))**: timestamp_granularities word or segment gives times, but it cannot be combined with language and there is no srt or vtt output.
- **OpenAI Speech to Text (OpenAI OpCo, LLC)**: timestamp_granularities word and srt or vtt exist only on whisper-1, which shuts on 2027-02-26; gpt-transcribe returns none of them.
- **Parakeet TDT 0.6B v3 (NVIDIA Corporation)**: Word and segment timestamps via timestamps=True in NeMo or durations in Transformers, but no srt or vtt and no speaker labels.
- **Qwen3-ASR-0.6B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check))**: Timestamps need a second model, Qwen3-ForcedAligner-0.6B, within 5 minutes of speech and 11 languages; no srt or vtt.
- **Qwen3-ASR-1.7B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check))**: Timestamps need a second model, Qwen3-ForcedAligner-0.6B, within 5 minutes of speech and 11 languages; no srt or vtt.
- **Smallest.ai Pulse STT (Smallest, Inc.)**: word_timestamps returns start and end per word with speaker labels, but no srt or vtt output.
- **Soniox Speech-to-Text (Soniox Inc.)**: Every token carries start and end times, but no srt or vtt output is documented.

- not this job: Cohere Transcribe 03-2026 (Cohere Inc.): No timestamps of any kind, per the docs.
- not this job: Giggy Speech to Text (Giggy): No word timestamps and no srt or vtt output, per the speech API docs.
- 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 no word-level timestamps.
- not this job: Ming-3.0-audio-flash (inclusionAI, Ant Group (named in the inclusion-ai.org footer; no registered company named)): No timestamps documented.
- not this job: Ming-3.0-audio-tiny (inclusionAI (Ant Group)): No timestamps documented.
- not this job: Nemotron 3.5 ASR Streaming 0.6B (NVIDIA Corporation): Timestamps are claimed in a subcard but not documented as an output; no srt or vtt.
- not this job: Underdog Bark 3.0 (Conway Research): No timestamps.
- not this job: Voxtral Mini 4B Realtime 2602 (Mistral AI (RCS Paris 952 418 325)): No timestamps, segments or subtitle output.

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

## What stops agents finishing

- no word timestamps on the current model, only on one being retired (OpenAI whisper-1 shuts 2027-02-26)
- no srt or vtt output (Groq, several open weights)
- timestamps need a second model (Qwen3-ASR needs Qwen3-ForcedAligner)
- timestamp drift on long files (acc-06 on AWS and Chirp)

## Around it

Before: media.download-video. After: video.burn-captions, video.cut-clips.

## Where this comes from

Weight 8 of 100 in Speech-to-text: 3 of 32 accounts (acc-05, acc-06, acc-12); word timestamps were also the stated reason two accounts stayed on Whisper despite its hallucinations

As data: https://www.anchorterminal.com/api/v1/use-cases/speech-to-text.subtitles-timestamps.json, or the MCP tool get_use_case with id speech-to-text.subtitles-timestamps. 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
