Audio transcription create

post

https://api.macpaw.com/v2/audio/transcriptions

Transcribes audio with an OpenAI or ElevenLabs model and returns the provider's response unchanged. Fields marked with a provider apply only to that provider's models. Provider docs: OpenAI, ElevenLabs.

The endpoint is synchronous: ElevenLabs webhook, webhook_id, webhook_metadata and cloud_storage_url are dropped without an error.

Recent Requests

Time	Status	User Agent
Retrieving recent requests…

Loading…

Body Params

model

string

required

Model id in provider/model format: openai/whisper-1, openai/gpt-4o-transcribe, openai/gpt-4o-mini-transcribe, elevenlabs/scribe_v1, elevenlabs/scribe_v2.

file

Audio or video file, up to 25 MB for OpenAI. Required unless source_url is sent; never together with it.

source_url

uri

ElevenLabs only — https URL of hosted audio or video. Required unless file is sent; never together with it.

language

string

Language of the audio; omit to auto-detect. OpenAI — ISO-639-1. ElevenLabs — ISO-639-1 or ISO-639-3.

temperature

number

Sampling temperature: 0 to 1 for OpenAI, 0 to 2 for ElevenLabs.

stream

boolean

Defaults to false

OpenAI only — return text/event-stream. Not supported by whisper-1.

prompt

string

OpenAI only — text that guides the style or continues a previous segment.

response_format

string

enum

Defaults to json

OpenAI only — output format. gpt-4o-transcribe and gpt-4o-mini-transcribe support json and text; gpt-4o-transcribe-diarize supports json, text and diarized_json; whisper-1 supports all but diarized_json.

Allowed:

timestamp_granularities[]

array of strings

OpenAI only — timestamp detail. Requires whisper-1 and response_format=verbose_json.

timestamp_granularities[]

Allowed:

include[]

array of strings

OpenAI only — logprobs returns token log probabilities. Works with gpt-4o-transcribe and gpt-4o-mini-transcribe and response_format=json.

include[]

Allowed:

chunking_strategy

string

OpenAI only — auto, or a JSON-encoded server_vad object. Required by gpt-4o-transcribe-diarize for audio longer than 30 seconds.

known_speaker_names[]

array of strings

OpenAI only (gpt-4o-transcribe-diarize) — up to 4 speaker names, matched by position to known_speaker_references[].

known_speaker_names[]

known_speaker_references[]

array of strings

OpenAI only (gpt-4o-transcribe-diarize) — audio samples of the known speakers as data URLs, 2 to 10 seconds each.

known_speaker_references[]

diarize

boolean

ElevenLabs only — label which speaker is talking.

num_speakers

integer

1 to 32

ElevenLabs only — maximum number of speakers in the audio.

diarization_threshold

number

ElevenLabs only — speaker separation threshold. Used with diarize=true when num_speakers is not set.

timestamps_granularity

string

enum

ElevenLabs only — timestamp detail in words[].

Allowed:

tag_audio_events

boolean

ElevenLabs only — tag non-speech sounds such as (laughter) in the transcript.

keyterms

array of strings

length ≤ 1000

ElevenLabs only — words or phrases to bias recognition toward. Each under 50 characters and at most 5 words. Billed at a surcharge by the provider.

keyterms

use_multi_channel

boolean

ElevenLabs only — transcribe each channel separately (up to 5). The response then has transcripts[] unless multichannel_output_style=combined.

multichannel_output_style

string

enum

ElevenLabs only — shape of a multi-channel response.

Allowed:

entity_detection

string

ElevenLabs only — all, a category (pii, phi, pci, other, offensive_language) or an entity type. Detected entities are returned in entities[].

entity_redaction

string

ElevenLabs only — entities to redact from the transcript; a subset of entity_detection.

entity_redaction_mode

string

enum

ElevenLabs only — how redacted entities are replaced.

Allowed:

no_verbatim

boolean

ElevenLabs only (scribe_v2) — remove filler words, false starts and non-speech sounds.

seed

integer

0 to 2147483647

ElevenLabs only — best-effort deterministic sampling.

file_format

string

enum

ElevenLabs only — pcm_s16le_16 for raw 16-bit, 16 kHz, mono PCM, which lowers latency.

Allowed:

Headers

string

enum

Defaults to application/json

Generated from available response content types

Allowed:

Responses

200The provider's transcription, unchanged. The shape depends on the provider, the model and response_format.

400Invalid form, a failed gateway check, or parameters the provider rejected (its message is in message).

401Unauthorized - Missing or invalid authentication token

402Insufficient credits

403Forbidden — authenticated but not allowed to access the requested model. Reported in the upstream error envelope rather than the gateway error envelope.