Transcription API

Audio Source

audio_file (required) — a URL to an audio file, or base64-encoded audio data. Payload limit: 10 MB for /run, 20 MB for /runsync.

Upload file Audio URL

Audio file (mp3, wav, ogg, flac, m4a, webm, mp4)

File will be base64-encoded and sent in the request payload. Keep source files under ~7 MB.

Audio URL (publicly accessible)

Passed directly to the API. No size limit.

Core Settings

language — ISO code (leave blank for auto-detect)

batch_size — parallelism level

Lower values use less GPU memory.

temperature — sampling randomness

initial_prompt — optional prompt for the first window

Language Detection

language_detection_min_prob

Minimum probability threshold

language_detection_max_tries

Maximum detection attempts

Voice Activity Detection

vad_onset

Threshold to start a speech segment

vad_offset

Threshold to end a speech segment

Diarization & Speaker Verification

diarization — assign speaker IDs

speaker_verification

min_speakers (blank = auto)

max_speakers (blank = auto)

speaker_samples — reference audio for verification

Provide a name and a publicly accessible URL for each known speaker.

Name

Audio URL

Output Options

align_output — word-level timestamp alignment

debug — include compute and memory info