API referenceTranscribe an audio file

Speech to text

Transcribe an audio file

Queue a long audio file (up to 2 hours) for background transcription. Send the audio as a multipart/form-data upload, optionally with recognition settings that tune how it is transcribed (priority words to boost, the model variant, and end-of-utterance silence). The request returns immediately with a job in queued status — nothing streams back and you do not keep the connection open. Poll the job endpoints to collect the transcription once it completes. Authenticate with the X-API-Key header.

POST

https://api.nabrah.ai/api/ext/stt/async-transcribe

curl --request POST \
  --url 'https://api.nabrah.ai/api/ext/stt/async-transcribe' \
  --header 'X-API-Key: <your-api-key>'

{
  "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "status": "queued",
  "project_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "result": null,
  "created_at": "2026-07-23T12:00:00Z",
  "updated_at": "2026-07-23T12:00:00Z",
  "filename": "recording.mp3"
}

Multipart form fields

Send the request as multipart/form-data (not JSON). Only audio_file is required; the recognition settings use the same names and defaults as the streaming WebSocket config.

audio_filefile · required

The audio file to transcribe. Maximum duration is 2 hours — longer files are rejected with a 422 error.

audio_file

priority_wordsstring · repeatable · optional

A word or phrase to bias recognition toward, such as product names, brands or branch names. Repeat the field once per word (for example -F priority_words=Nabrah -F priority_words=Olaya). Up to 100 words; more returns a 400 error.

priority_words

priority_words_strengthnumber · default: 0.5

How strongly to bias toward priority_words. 0.5 works well for a relatively small vocabulary; the recommended range for broader biasing is 20-100. A value that is not a number returns a 400 error.

priority_words_strength

recognition_modelstring · default: server-configured model

Model variant for the job. Leave it out to use the default model, which is the most accurate.

recognition_model

end_of_utterance_silence_msinteger · default: -1

Trailing silence in milliseconds that ends an utterance. -1 keeps the recognizer's own defaults; only a positive value overrides. A value that is not an integer returns a 400 error.

end_of_utterance_silence_ms

The settings a job was queued with are returned on the job as recognition_config, so you can confirm what was applied.

Async flow

Transcription runs in the background; this request only creates the job.

1 · Upload

This endpoint stores the audio, queues a transcription job, and returns the job immediately with status: "queued" and result: null.

2 · Track

Poll GET /api/ext/stt/jobs/{job_id} (or list jobs with GET /api/ext/stt/jobs). The status moves queued → in_progress → completed, or failed if transcription could not finish.

3 · Result

Once status is completed, the job's result object holds the transcription: text_result, audio_duration, and timestamps.

text_resultaudio_durationtimestamps

Authorization

X-API-Keystringheaderrequired

Your workspace API key. Create one in the dashboard under API keys.

Body

multipart/form-data

Response

200

Successful Response

application/json

422

Validation Error

application/json

API referenceTranscribe an audio file

Speech to text

Transcribe an audio file

POST

https://api.nabrah.ai/api/ext/stt/async-transcribe

Transcribe an audio file

curl --request POST \
  --url 'https://api.nabrah.ai/api/ext/stt/async-transcribe' \
  --header 'X-API-Key: <your-api-key>'

{
  "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "status": "queued",
  "project_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "result": null,
  "created_at": "2026-07-23T12:00:00Z",
  "updated_at": "2026-07-23T12:00:00Z",
  "filename": "recording.mp3"
}

Multipart form fields

Send the request as multipart/form-data (not JSON). Only audio_file is required; the recognition settings use the same names and defaults as the streaming WebSocket config.

audio_filefile · required

The audio file to transcribe. Maximum duration is 2 hours — longer files are rejected with a 422 error.

audio_file

priority_wordsstring · repeatable · optional

priority_words

priority_words_strengthnumber · default: 0.5

How strongly to bias toward priority_words. 0.5 works well for a relatively small vocabulary; the recommended range for broader biasing is 20-100. A value that is not a number returns a 400 error.

priority_words_strength

recognition_modelstring · default: server-configured model

Model variant for the job. Leave it out to use the default model, which is the most accurate.

recognition_model

end_of_utterance_silence_msinteger · default: -1

Trailing silence in milliseconds that ends an utterance. -1 keeps the recognizer's own defaults; only a positive value overrides. A value that is not an integer returns a 400 error.

end_of_utterance_silence_ms

The settings a job was queued with are returned on the job as recognition_config, so you can confirm what was applied.

Async flow

Transcription runs in the background; this request only creates the job.

1 · Upload

This endpoint stores the audio, queues a transcription job, and returns the job immediately with status: "queued" and result: null.

2 · Track

Poll GET /api/ext/stt/jobs/{job_id} (or list jobs with GET /api/ext/stt/jobs). The status moves queued → in_progress → completed, or failed if transcription could not finish.

3 · Result

Once status is completed, the job's result object holds the transcription: text_result, audio_duration, and timestamps.

text_resultaudio_durationtimestamps

Authorization

X-API-Keystringheaderrequired

Your workspace API key. Create one in the dashboard under API keys.

Body

multipart/form-data

Response

200

Successful Response

application/json

422

Validation Error

application/json