Skip to main content
POST
Transcribe a file

Authorizations

Authorization
string
header
required

Your API key

Body

multipart/form-data
file
file

The audio. Any common format (WAV, MP3, ...).

model
string

umay (most accurate; files), or an OpenAI name such as whisper-1, gpt-4o-transcribe, gpt-4o-transcribe-diarize.

Example:

"umay"

language
enum<string>

Language of the speech.

Available options:
tr,
en,
de,
fr,
it,
ru,
es
Example:

"tr"

normalize
boolean
default:true

Inverse text normalisation on the final text: "bin iki yüz elli lira" becomes "1.250 TL". false returns the spoken words.

dialect_correction
boolean
default:true

Turkish, with normalize=true: regional and colloquial forms in standard Turkish.

place_correction
boolean
default:true

Turkish, with normalize=true: misheard or split place names corrected.

abbreviations
boolean
default:false

Turkish, with normalize=true: false writes units out ("43 dakika"), true abbreviates ("43 dk").

mode
enum<string>
default:fast

normal: files with model=umay use the language model where the model has one. fast: never.

Available options:
fast,
normal
hint_context
string

Built-in hint lists (Turkish): mugla, izmir, manisa, aydin, denizli; comma-separated for several.

hints
string

Own phrases (names, places) to favour, comma or newline separated; phrase :3 gives one phrase its own boost.

hints_score
number
default:2

Boost for phrases without their own.

gender_detection
boolean
default:false

Gender of the speaker(s) with a confidence percent (50-100).

diarize
boolean
default:false

Speaker label per segment (S1, S2, ...).

max_speakers
integer
default:8

Upper bound for diarization.

speaker_lock
boolean
default:false

Only the first speaker's speech becomes text; other voices come as ignored.

lock_threshold
number

With speaker_lock: the similarity a segment needs to count as S1 (0-1). Default 0.5, or 0.55 with reference_voiceprint.

voice_isolation
boolean
default:false

Removes other voices before recognition (tr, de, fr, it, es, ru); switches on only after another voice is heard.

reference_voiceprint
string

Id or token of a stored voiceprint: with voice_isolation, speaker_lock or diarize, speaker S1 is that person from the first second.

response_format
enum<string>
default:json

With OpenAI model names.

Available options:
json,
text,
srt,
vtt,
verbose_json,
diarized_json
webhook_url
string

https:// URL that receives the result (POST, JSON). With it the request answers 202 at once. Must resolve to a public address.

webhook_secret
string

Optional, 1-256 characters: the webhook is signed with it (X-Mihu-Signature).

webhook_metadata
string

Optional, up to 4 KB: echoed back as metadata.

file_url
string

Instead of file: https:// URL the server downloads the audio from (up to 1 GB, 10 minutes, public addresses only).

Response

Transcription