Elevenlabs Scribe V3 API Documentation
Playground
Try it on WaveSpeedAI!ElevenLabs Scribe V3 Speech-to-Text transcribes recorded audio with automatic language detection, speaker labels, word-level timestamps, number formatting, and multilingual language hints for transcription, subtitles, captions, and meetings. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.
Features
ElevenLabs Scribe V3 transcribes audio into structured text with automatic language detection, optional speaker diarization, word-level timestamps, audio-event tagging, number formatting, and language hints for multilingual recordings.
Use optional keyterms to provide recognition hints for names, technical terms, brands, or other vocabulary that may be difficult to recognize from audio alone.
Why Choose This?
- Speech-to-text transcription — Convert recorded speech into structured transcript data.
- Automatic language detection — Let the model detect the spoken language automatically, or provide a language code when known.
- Speaker diarization — Enable
diarizeto label different speakers in multi-speaker recordings. - Word-level timestamps — Receive timing information for individual words for subtitle, editing, alignment, and search workflows.
- Audio-event tagging — Include detected non-speech events such as laughter when
tag_audio_eventsis enabled. - Keyterm guidance — Provide up to
100recognition hints for names, brands, technical vocabulary, or domain-specific terms. - Number formatting — Choose whether numbers are written as digits or words.
- Multilingual language hints — Provide additional language codes for recordings containing more than one language.
Parameters
| Parameter | Required | Description |
|---|---|---|
| audio | Yes | Audio input provided by upload or public direct file URL. |
| language_code | No | ISO-639-1 or ISO-639-3 language code. Omit for automatic language detection. |
| secondary_language_codes | No | Up to 100 additional ISO-639-1 or ISO-639-3 language codes expected alongside the main language. Omit when not needed. |
| numbers_format | No | Write numbers as digits (42) or words (forty-two). Omit to use the model default. |
| diarize | No | Identify and label different speakers in the recording. Default: false. |
| tag_audio_events | No | Include detected audio events such as laughter in the transcription result. Default: true. |
| keyterms | No | Recognition hints for important vocabulary. Supports up to 100 items; each term must be under 50 characters and contain at most 5 words. Additional pricing applies when any keyterms are provided. |
How to Use
- Provide the audio — Upload a recording or supply a direct public audio-file URL.
- Set language optional — Provide
language_codewhen the spoken language is known, or omit it for automatic detection. - Enable diarization optional — Turn on
diarizewhen speaker labels are needed. - Configure audio events optional — Keep
tag_audio_eventsenabled when non-speech events should be included. - Add keyterms optional — Supply names, brands, technical vocabulary, or other recognition hints when needed.
- Set multilingual hints and number formatting optional — Add
secondary_language_codesand choosenumbers_formatwhen needed. - Submit — Process the recording and retrieve the transcript result.
Pricing
Pricing is based on the processed input audio duration.
| Configuration | Price per Input Minute |
|---|---|
| Standard transcription | $0.01 |
| With one or more keyterms | $0.012 |
Each request has a minimum billable duration of 10 seconds. Audio longer than 10 seconds is charged proportionally to its duration rather than rounded up to full minutes.
Example Costs
| Audio Duration | Standard | With Keyterms |
|---|---|---|
| 30s | $0.005 | $0.006 |
| 1 min | $0.01 | $0.012 |
| 10 min | $0.10 | $0.12 |
| 30 min | $0.30 | $0.36 |
| 60 min | $0.60 | $0.72 |
Only the first 60 minutes of input audio are processed and billed.
Providing one or more keyterms applies a 1.2× multiplier to the transcription price. The number of keyterms does not add separate per-item charges.
Best Use Cases
- Interviews — Transcribe conversations with optional speaker labels.
- Meetings — Create transcripts with speaker labels and word timing.
- Podcasts — Convert spoken episodes into text for publishing, indexing, or subtitle workflows.
- Subtitle preparation — Use word-level timestamps in caption and subtitle workflows.
- Technical recordings — Use
keytermsto provide recognition hints for specialized terminology. - Content search and indexing — Convert audio archives into structured text for downstream processing.
Pro Tips
- Enable
diarizewhen multiple people speak in the same recording. - Provide
language_codewhen the spoken language is already known. - Use
keytermsfor uncommon names, brands, acronyms, product names, or technical vocabulary. - Keep keyterms short and specific instead of using full explanatory sentences.
- Clear speech and low background noise are recommended.
- Supply a direct audio-file URL rather than a webpage containing an embedded audio player.
Notes
- Each request is billed for at least
10seconds. This billing minimum does not extend or repeat the input audio. - Only the first
60minutes of input audio are processed and billed. keytermssupports up to100items.- Each keyterm must be under
50characters and contain no more than5words. It cannot include<,>,{,},[,], or a backslash. - Keyterms provide recognition guidance but do not guarantee exact spelling.
- Public URLs must point directly to an accessible audio file.
Related Models
- ElevenLabs Scribe V2 — Transcribe audio with language detection and word-level timestamps.
- ElevenLabs Forced Alignment — Align an existing transcript with audio and return detailed timing information.
Authentication
For authentication details, please refer to the Authentication Guide.
API Endpoints
Submit Task & Query Result
set -euo pipefail
export WAVESPEED_API_KEY="your-api-key"
REQUEST_BODY=$(cat <<'JSON'
{
"audio": "https://interactive-examples.mdn.mozilla.net/media/cc0-audio/t-rex-roar.mp3",
"numbers_format": "digits",
"diarize": false,
"tag_audio_events": true,
"keyterms": []
}
JSON
)
# 1. Submit the prediction once. Never retry this request automatically:
# a retry after an unclear failure can create a second, billed task.
CHECK='The task may already exist. Check History or the predictions list before submitting again.'
if ! SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
--connect-timeout 10 --max-time 60 \
-X POST "https://api.wavespeed.ai/api/v3/elevenlabs/scribe-v3" \
-H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
-H "Content-Type: application/json" \
-d "${REQUEST_BODY}"); then
printf 'Submit failed. %s\n' "${CHECK}" >&2
exit 1
fi
TASK=$(printf '%s' "${SUBMIT_RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
PREDICTION_ID=$(printf '%s' "${TASK}" | jq -r '.id // empty')
if [ -z "${PREDICTION_ID}" ]; then
printf 'Submission response did not contain a prediction id. %s\n' "${CHECK}" >&2
exit 1
fi
RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"
# 2. Poll until the prediction reaches a final status.
BODY_FILE=$(mktemp)
trap 'rm -f "${BODY_FILE}"' EXIT
while true; do
# Transient errors (429, 5xx) are retried, honoring Retry-After.
HTTP_CODE=$(curl --silent --show-error --fail-with-body \
--connect-timeout 10 --max-time 60 --retry 5 --retry-delay 2 \
--output "${BODY_FILE}" --write-out '%{http_code}' \
"${RESULT_URL}" \
-H "Authorization: Bearer ${WAVESPEED_API_KEY}" || true)
case "${HTTP_CODE}" in
2??) ;;
000|429|5??) # dropped connection or temporary error: keep polling
printf 'Task %s: HTTP %s, polling again in 10s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
sleep 10; continue ;;
*) printf 'Task %s: HTTP %s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
cat "${BODY_FILE}" >&2; exit 1 ;;
esac
RESULT=$(jq 'if type == "object" and has("data") then .data else . end' "${BODY_FILE}")
STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')
case "${STATUS}" in
completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
failed|cancelled|timeout|deleted)
printf 'Task %s ended with status %s\n' "${PREDICTION_ID}" "${STATUS}" >&2
printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
*) sleep 2 ;;
esac
doneRetry temporary errors only while polling, never the submit request. Errors, retries and timeouts
Parameters
Task Submission Parameters
Request Parameters
| Parameter | Type | Required | Default | Range | Description |
|---|---|---|---|---|---|
| audio | string | Yes | - | 1 ~ unlimited characters · pattern: ^https?:// | Public direct URL of the audio file. Upload a file or supply a URL. |
| language_code | string | No | - | pattern: ^[a-z]{2,3}$ | ISO-639-1 or ISO-639-3 language code. Omit for automatic detection. |
| secondary_language_codes | array<string> | No | - | 0 ~ 100 items | Up to 100 additional ISO-639-1 or ISO-639-3 language codes expected in the audio, alongside language_code. Omit when not needed. |
| numbers_format | string | No | - | digits, words | Write numbers as digits (42) or words (forty-two). Omit to use the model default. |
| diarize | boolean | No | false | - | Identify different speakers. |
| tag_audio_events | boolean | No | true | - | Include audio events such as laughter. |
| keyterms | array<string> | No | [] | 0 ~ 100 items · each item: 1 ~ 49 characters · pattern: ^[^\s<>{}\[\]\\]+(?:\s+[^\s<>{}\[\]\\]+){0,4}$ | Up to 100 recognition hints, each under 50 characters and at most 5 words. Additional charge applies. |
Every model also accepts the global request options enable_sync_mode and enable_base64_output; see their pages for details.
Response Parameters
| Parameter | Type | Description |
|---|---|---|
| code | integer | HTTP status code (e.g., 200 for success) |
| message | string | Status message (e.g., “success”) |
| data.id | string | Unique identifier for the prediction, Task Id |
| data.model | string | Model ID used for the prediction |
| data.outputs | array | Output values, usually URL strings; some models return text strings or structured result objects (empty when status is not completed) |
| data.urls | object | Object containing related API endpoints |
| data.status | string | Task status. completed is successful; failed, cancelled, timeout, and deleted are failure terminal statuses. |
| data.created_at | string | ISO timestamp of when the request was created (e.g., “2023-04-01T12:34:56.789Z”) |
| data.error | string | Error message (empty if no error occurred) |
| data.timings | object | Object containing timing details |
| data.timings.inference | integer | Inference time in milliseconds |
Result Request Parameters
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
| id | string | Yes | - | Task ID |
Result Response Parameters
| Parameter | Type | Description |
|---|---|---|
| code | integer | HTTP status code (e.g., 200 for success) |
| message | string | Status message (e.g., “success”) |
| data | object | The prediction data object containing all details |
| data.id | string | Unique identifier for the prediction |
| data.model | string | Model ID used for the prediction |
| data.outputs | array<string | object> | Array of generated outputs (empty when status is not completed). Items are usually URL strings, but may be text strings or structured result objects, depending on the model. |
| data.urls | object | Object containing related API endpoints |
| data.status | string | Status: completed is successful; failed, cancelled, timeout, and deleted are failure terminal statuses |
| data.created_at | string | ISO timestamp of when the request was created |
| data.error | string | Error message (empty if no error occurred) |
| data.timings | object | Object containing timing details |
| data.timings.inference | integer | Inference time in milliseconds |