Speech to text

Transcription is available on its own, without building a voice agent. Send audio in, get text back.

https://stt.omnia-voice.com

Authenticate with the same key you use for the rest of the API, in the X-API-Key header.

	Endpoint	Use it for
File	`POST /transcribe`	Recordings up to 60 seconds. Synchronous — the text comes back in the response.
Long audio	`POST /batch`	Meetings, calls, podcasts, up to 8 hours. Asynchronous — poll for the result.
Streaming	`wss://.../stream` and `wss://.../v2/stream`	Live audio, as it is being spoken.

Languages

Every endpoint accepts a languages option: BCP-47 codes, up to 4, or auto to detect the language. On the REST endpoints pass it as the X-Languages header or ?languages= query parameter. The default is Finnish and Swedish (fi-FI,sv-SE) except on /v2/stream, which auto-detects.

-H "X-Languages: et-EE,ru-RU"

Supported languages

GET /languages lists every supported code with a status of stable or preview. No key needed. Add ?codes= to check specific ones before you build on them.

curl "https://stt.omnia-voice.com/languages?codes=et-EE,ru-RU"

{
  "codes": ["et-EE", "ru-RU"],
  "maxPerRequest": 4,
  "engines": {
    "v1": {
      "et-EE": { "supported": true, "status": "preview", "name": "Estonian (Estonia)" },
      "ru-RU": { "supported": true, "status": "stable", "name": "Russian (Russia)" }
    },
    "v2": {
      "et-EE": { "supported": true, "status": "preview", "name": "Estonian" },
      "ru-RU": { "supported": true, "status": "stable", "name": "Russian" }
    }
  }
}

v1 covers /transcribe, /batch and /stream; v2 covers /v2/stream.

Transcribe a file

Up to 10 MB and 60 seconds of audio. For anything longer, use /batch.

curl -X POST "https://stt.omnia-voice.com/transcribe" \
  -H "X-API-Key: $OMNIA_API_KEY" \
  -H "Content-Type: audio/mpeg" \
  -H "X-Languages: fi-FI,sv-SE" \
  --data-binary @audio.mp3

import { readFile } from "node:fs/promises";

const audio = await readFile("audio.mp3");

const res = await fetch("https://stt.omnia-voice.com/transcribe", {
  method: "POST",
  headers: {
    "X-API-Key": process.env.OMNIA_API_KEY,
    "Content-Type": "audio/mpeg",
    "X-Languages": "fi-FI,sv-SE",
  },
  body: audio,
});

const { transcript } = await res.json();
console.log(transcript);

import os, requests

with open("audio.mp3", "rb") as f:
    res = requests.post(
        "https://stt.omnia-voice.com/transcribe",
        headers={
            "X-API-Key": os.environ["OMNIA_API_KEY"],
            "Content-Type": "audio/mpeg",
            "X-Languages": "fi-FI,sv-SE",
        },
        data=f,
    )

print(res.json()["transcript"])

Response

{
  "transcript": "Moi, Kati tässä. Oisko sulla hetki aikaa?",
  "segments": [
    { "transcript": "Moi, Kati tässä. Oisko sulla hetki aikaa?", "confidence": 0, "language": "fi-FI", "isFinal": true }
  ],
  "audioSize": 184320,
  "estimatedDurationMs": 5760,
  "processingTime": 1480,
  "sessionId": "stt_1788786506604_8fd6cc541a02b620"
}

string

The full text.

array

The text split by the engine, each with the detected language.

integer

Audio length in milliseconds, estimated from the file. This is what you are billed on.

integer

How long recognition took, in milliseconds.

Transcribe long audio

POST /batch takes the same headers as /transcribe, accepts up to 500 MB and 8 hours, and returns an operation to poll. Add X-Diarization: true to get a speakerTurns array saying who said what.

# Start
curl -X POST "https://stt.omnia-voice.com/batch" \
  -H "X-API-Key: $OMNIA_API_KEY" \
  -H "Content-Type: audio/mp4" \
  -H "X-Languages: auto" \
  --data-binary @meeting.m4a
# → { "operationId": "batch_…", "status": "processing", "pollUrl": "/batch/batch_…" }

# Poll every 10–30 s until status is "completed"
curl "https://stt.omnia-voice.com/batch/batch_…" -H "X-API-Key: $OMNIA_API_KEY"
# → { "status": "completed", "transcript": "…", "segments": […], "durationMs": 3600000 }

Results are kept for 24 hours. Direct uploads are capped at 32 MB by the platform; for larger files call POST /batch/init, PUT the audio to the returned uploadUrl, then POST /batch/{operationId}/start. The full flow is in the API reference at https://stt.omnia-voice.com/.

Audio formats

Send the matching Content-Type.

Format	`/transcribe`	`/batch`
MP3 `audio/mpeg`	✓	✓
WAV `audio/wav`	✓	✓
FLAC `audio/flac`	✓	✓
OGG Opus `audio/ogg`	✓	✓
WebM Opus `audio/webm`	✓	✓
MP4 / M4A `audio/mp4`		✓
Raw PCM `audio/x-raw` (16 kHz, 16-bit, mono)	✓	✓

Quality in beats quality out. 16 kHz mono is plenty for speech and keeps uploads small; a noisy 48 kHz stereo recording transcribes worse than a clean 16 kHz one, not better.

Billing

4 credits per minute ($0.04), rounded up, one-minute minimum. See Credits.

Live streaming

Transcribe audio as it is spoken, with interim results.

Get started

Core concepts

Tools

Telephony

Apps & SDKs

Guides

Examples

Help

Speech to text

Languages

Supported languages

Transcribe a file

Response

Transcribe long audio

Audio formats

Billing

Live streaming

​Languages

​Supported languages

​Transcribe a file

​Response

​Transcribe long audio

​Audio formats

​Billing

Live streaming

Languages

Supported languages

Transcribe a file

Response

Transcribe long audio

Audio formats

Billing