by ElevenLabs

Word-level transcripts with speaker labels and sound events in 102 languages

elevenlabs/scribe-v3/speech-to-text
  • 112 languages
  • Speech to Text

About ElevenLabs Scribe V3

Word-level transcripts with speaker labels and sound events in 102 languages.

Call it with the model name elevenlabs/scribe-v3/speech-to-text on POST /jobs/createTask, from $0.0002.

Introduction

Turn a recording into text you can review

ElevenLabs Scribe V3 transcribes recorded speech from audio_url. Use it to prepare an interview transcript, recover the wording of a spoken note or locate passages in a recorded talk. The response includes text and word timings, giving you a way to move between the transcript and its source while editing.

ElevenLabs describes Scribe V3 as its recorded-audio model and Scribe V3 Live as its realtime model in the launch announcement. Choose this endpoint when you have a recording ready to transcribe. A recording can run up to one hour.

Set up the transcription request

Start with the URL of the recording. Choose the language context and presentation options according to the transcript you need: a conversation organized by speaker, a draft with sound markers, or prose with a consistent written form for numbers.

ParameterRequiredWhat to supply
audio_urlYesA nonempty recording URL for audio of up to one hour.
languageNoAn ISO 639 code from the supported list, such as en, zh, ja or es. Leave empty for detection.
language_hintsNoAn array of other expected language codes, up to 100 entries; one code per row in the form.
numbers_formatNodigits or words, depending on how you want spoken numbers written.
diarizeNoA boolean requesting speaker labels. Defaults to false.
tag_audio_eventsNoA boolean requesting sound markers such as laughter. Defaults to true.

Give the recording useful language context

Set language when you know the spoken language. For a recording with other expected languages, add their codes through language_hints. Choose codes that match the speech you expect to hear, and leave the main language empty when you want detection. An interview with a known main language and occasional passages in another language is a practical reason to use both controls.

ElevenLabs says Scribe transcribes speech in its spoken language rather than translating it; see the number and language guide. Plan your review around the language of the recording, including quotations or names that may need a listener familiar with that language.

Choose how numbers and speaking turns appear

Use numbers_format set to digits for a transcript where you will check quantities, or words for a prose draft. Match the setting to the next editing step, then listen back to values that matter to the final document. A consistent number style makes a transcript easier to scan while you review figures in context.

Enable diarize when speaker labels will help organize a conversation. Check each turn against the audio before attaching a quotation to a named participant. Keep tag_audio_events enabled if laughter and similar sound markers help convey the exchange; turn it off if those markers would interrupt the reading format you need.

Prepare the source and review the result

Listen to the recording before submitting it, noting passages with speaker changes, unfamiliar names or important numbers. Keep the source available alongside the transcript so you can check those passages directly. For a short spoken note, compare the returned wording with the message you intended to capture. For an interview, review turns and quotations together.

Use word timings to find passages in the audio while correcting the text. When preparing captions, review the wording and timing in your editing workflow before publication. Preserve the connection between the transcript and its source as you revise so a colleague can check a quotation or figure without searching the entire recording.

A short English note, transcribed

Our November 18, 2026 run used a 2.35-second English recording. The next-step-note example returned “The next step is ready. Please continue.” The response reported language code eng and included start and end times for every word and space. The returned text captures the progress note, and the timings locate its words in the recording.

Compare related transcription families

Whisper 1 and Grok STT are related speech-to-text families to consider for the same work. Compare their transcripts using a recording that resembles your interviews, talks or notes.

For a project that needs a new sound effect, explore ElevenLabs Sound Effects V3. Choose Scribe V3 when the next step is recovering words from an existing recording and reviewing them against the audio.

Request parameters

ParameterRequiredDescription
audio_urlYesThe voice or music track to use.
diarizeNoLabel who is speaking. Default: false
languageNoThe language of the speech. en · zh · ja · ko · es · fr · de · pt · ru · ar · hi · afr · amh · ara · hye · asm · ast · aze · bel · ben · bos · bul · mya · yue · cat · ceb · nya · hrv · ces · dan · nld · eng · est · fil · fin · fra · ful · glg · lug · kat · deu · ell · guj · hau · heb · hin · hun · isl · ibo · ind · gle · ita · jpn · jav · kea · kan · kaz · khm · kor · kur · kir · lao · lav · lin · lit · luo · ltz · mkd · msa · mal · mlt · zho · mri · mar · mon · nep · nso · nor · oci · ori · pus · fas · pol · por · pan · ron · rus · srp · sna · snd · sin · slk · slv · som · spa · swa · swe · tam · tgk · tel · tha · tur · ukr · umb · urd · uzb · vie · cym · wol · xho · yor · zul
language_hintsNoLikely spoken language codes, one per row.
numbers_formatNoWrite numbers as digits or as words. digits · words
tag_audio_eventsNoMark sounds such as laughter in the transcript. Default: true

Pricing

Prices are in USD and charged to your PopNow balance. When a task is accepted, we reserve the maximum quoted amount from your API balance. On completion, we charge for actual usage, never more than the hold, and release the remainder. If the task fails, the hold is released.

OptionUSD per second
Standard$0.0002

Send the same request body to POST /jobs/quote to see the estimated and maximum USD charge before running. A quote does not run the task or charge your balance.

API quick start

Call the ElevenLabs Scribe V3 API in three steps

  1. Create an API key. Sign in to PopNow and create a key in the API console. Keys start with sk-popnow-. API calls are billed in USD from your PopNow balance.
  2. Send the request. POST /jobs/createTask with model elevenlabs/scribe-v3/speech-to-text and an input object that follows the parameter table. The response returns a task ID straight away.
  3. Get the result. Poll GET /jobs/recordInfo with the task ID, or pass a callBackUrl and we post the result to it. Download the output links when the task succeeds.
# 1. Create the task. Keep the Idempotency-Key if you retry after a network error.
export IDEMPOTENCY_KEY="$(uuidgen)"
curl -X POST https://api.popnow.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $POPNOW_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $IDEMPOTENCY_KEY" \
  -d '{
  "model": "elevenlabs/scribe-v3/speech-to-text",
  "input": {
    "audio_url": "https://media.popnow.ai/shortfilm/voices/Aoede-46e4c427.mp3"
  }
}'

# 2. Copy data.taskId from the response, then poll until state is succeeded, failed, canceled or expired.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.popnow.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $POPNOW_API_KEY"

Replace the prompt and any links with your own. The API tab has the response, polling and callback reference.

ElevenLabs Scribe V3 API questions

How much does the ElevenLabs Scribe V3 API cost?

ElevenLabs Scribe V3 costs $0.0002 per second. API usage is charged in USD to your PopNow balance. When a task is accepted, we reserve the maximum quoted amount from your API balance. On completion, we charge for actual usage, never more than the hold, and release the remainder. If the task fails, the hold is released.

What is the model name for ElevenLabs Scribe V3 in the API?

Put elevenlabs/scribe-v3/speech-to-text for Speech to Text in the model field of POST /jobs/createTask.

What does ElevenLabs Scribe V3 return?

It returns a transcript of recorded speech with word timings. In our 2.35-second English run, the response contained the text, detected language code eng, and start and end times for every word and space. Speaker labels and audio event tags are available through separate controls.

How do I use the ElevenLabs Scribe V3 API?

Provide the recording URL in audio_url. Add a language code if you know the spoken language, then choose whether to label speakers, tag audio events or format numbers. Only transcribe recordings you have the right to use.

Should I set a language or use language hints?

Set language when you know the spoken language. Leave it empty for detection, and use language_hints for other languages expected in the recording, with up to 100 codes. ElevenLabs says Scribe transcribes speech in its spoken language rather than translating it; see its [number and language guide](https://elevenlabs.io/docs/eleven-api/guides/how-to/speech-to-text/batch/number-and-language-options).

Can Scribe V3 label different speakers?

Yes, enable diarize to request speaker labels. It is off by default, so choose it deliberately for an interview or meeting transcript. Check the labels against the recording before assigning quotations to speakers.

Why might laughter appear in a transcript?

Audio event tagging is on by default. The tag_audio_events control marks sounds such as laughter in the transcript; switch it off when those markers are not wanted.

What are the limits of the ElevenLabs Scribe V3 API?

Largest upload: 90 MB; Languages: 112. Requests outside these limits are rejected before anything is charged.

Can I try ElevenLabs Scribe V3 without writing code?

Run ElevenLabs Scribe V3 in the Playground with your API USD balance, without an API key. Each run appears in your developer history with its matching code.

All ElevenLabs models

More about ElevenLabs Scribe V3