Developers / Docs / Speech to text

Speech to text

Transcribe recordings, calls and meetings with Whisper large-v3-turbo. Get plain text, timestamped segments, or ready-made subtitles.

API reference: Audio transcriptions

POST https://api.ark-labs.cloud/api/v1/audio/transcriptions

Transcribe a file

Send the file as multipart/form-data. Common formats work: MP3, WAV, OGG, FLAC, M4A and the audio track of most video files.

with open("meeting.mp3", "rb") as audio:
    transcript = client.audio.transcriptions.create(
        model="arklabsai/whisper-large-v3-turbo",
        file=audio,
        language="pl",
    )
print(transcript.text)

import fs from "node:fs";

const transcript = await client.audio.transcriptions.create({
  model: "arklabsai/whisper-large-v3-turbo",
  file: fs.createReadStream("meeting.mp3"),
  language: "pl",
});
console.log(transcript.text);

curl https://api.ark-labs.cloud/api/v1/audio/transcriptions \
  -H "Authorization: Bearer $ARK_API_KEY" \
  -F model=arklabsai/whisper-large-v3-turbo \
  -F file=@meeting.mp3 \
  -F language=pl

Response

{
  "text": "Dzień dobry, zaczynamy przegląd kwartalny…",
  "usage": { "type": "duration", "seconds": 1834.2, "credits_used": 918, "credit_balance": 49069 }
}

Parameters

filefilerequired

The audio to transcribe.

modelstringrequired

arklabsai/whisper-large-v3-turbo.

languagestringoptional

The spoken language as an ISO-639-1 code, such as pl, en or de. Set it whenever you know the language: ARK runs no separate language detection. An unsupported language returns 400.

response_formatstringoptional

json (default), text, verbose_json, srt or vtt.

temperaturenumber 0–1optional

Sampling temperature. Defaults to 0.

timestamp_granularitiesarrayaccepted, no effect

Accepted for compatibility. verbose_json always returns segment-level timestamps; word-level timestamps are not available.

Response formats

Format	You receive
`json`	`{"text": …, "usage": …}`
`verbose_json`	The text plus `duration`, `language` and `segments` with start and end times in seconds.
`text`	The transcript as plain text.
`srt`	SubRip subtitles.
`vtt`	WebVTT subtitles.

curl https://api.ark-labs.cloud/api/v1/audio/transcriptions \
  -H "Authorization: Bearer $ARK_API_KEY" \
  -F model=arklabsai/whisper-large-v3-turbo \
  -F file=@interview.wav \
  -F language=en \
  -F response_format=srt \
  -o interview.srt

srt = client.audio.transcriptions.create(
    model="arklabsai/whisper-large-v3-turbo",
    file=open("interview.wav", "rb"),
    language="en",
    response_format="srt",
)
open("interview.srt", "w").write(srt)

interview.srt

1
00:00:00,000 --> 00:00:04,120
Thanks for joining us today.

2
00:00:04,600 --> 00:00:09,880
Let's start with how your team moved its workloads to Europe.

How audio is processed

ARK finds the speech in the recording, transcribes each stretch of speech separately and joins the results, so long recordings with pauses are handled well. A stretch that cannot be made out appears as [inaudible] in subtitles.

Billing

Transcription is billed by the duration of the audio file, silence included, at a price per audio hour. usage.seconds in json and verbose_json responses shows the billed duration. See Pricing.