Skip to main content
POST
Transcribe a completed audio file

Authorizations

Authorization
string
header
required

Gateway API key (imp-rt-...) from POST /v1/accounts or /v1/keys.

Body

multipart/form-data
model
string
required
Example:

"openai/whisper-large-v3-turbo"

file
file
required

Completed audio file in a format supported by the selected model.

Maximum string length: 26214400
prompt
string

Unstructured recording context for OpenAI and Groq models. xAI uses keyterm instead. Unsupported for ElevenLabs Scribe.

keywords[]
string[]

Literal terms expected in the recording. Repeat this multipart field for multiple terms.

languages[]
string[]

Expected language codes for OpenAI gpt-transcribe. Repeat this field for multiple languages.

language
string

Single language code for Groq Whisper and xAI STT. Unsupported for OpenAI gpt-transcribe. ElevenLabs Scribe receives it as language_code.

language_code
string

ElevenLabs Scribe's native name for language. Send language or language_code, not both.

diarize
boolean

ElevenLabs Scribe speaker labels. Other ElevenLabs controls (num_speakers, diarization_threshold, tag_audio_events, timestamps_granularity, file_format, additional_formats, use_multi_channel, multichannel_output_style, enable_logging) also pass through.

keyterm
string[]

xAI transcription vocabulary hints. Repeat this multipart field for multiple terms.

temperature
number

Groq Whisper sampling temperature.

Required range: 0 <= x <= 1
timestamp_granularities[]
enum<string>[]

Groq timestamps; requires response_format=verbose_json.

Available options:
word,
segment
response_format
enum<string>
default:json

gpt-transcribe accepts json or text; Groq and xAI also accept verbose_json. ElevenLabs Scribe returns its own JSON for json and verbose_json, and plain text for text.

Available options:
json,
text,
verbose_json
stream
boolean
default:false

Response

A completed transcript object, or transcript events when stream is true.

text
string
required
languages
object[]
usage
object

Present when OpenAI reports duration usage.