Turn speech into text—on your hardware.

A focused, OpenAI-compatible transcription service for Swedish and English. Endpointed audio, Silero VAD, pinned models, no cloud account.

Try your audio Read the API docs

Languages: Swedish + English
Runtime: CPU · amd64 + arm64
Privacy: Local by default

live request complete

sv-short.wav 00:05.01

TRANSCRIPTSV · SWEDISH SMALL

Hej! Hur mår du? Jag talar svenska.

00:00.19Hej! Hur mår du?
00:02.75Jag talar svenska.

Built for the request path

Predictable where it matters.

The gateway keeps native inference behind a small, stable HTTP boundary. Clients choose a language; the service handles routing, endpointing, and normalization.

Explicit language routing

Short utterances are ambiguous. Choose sv or en and always reach the intended model.

Deterministic

Silence-aware output

Silero VAD filters non-speech while the gateway normalizes unstable worker details into one public schema.

VAD 6.2

Local, pinned, portable

Checksummed artifacts and native multi-architecture images make the same stack reproducible across machines.

Docker Compose

Studio

Hear it. Drop it. Inspect every segment.

Record from your microphone or upload an existing file. Change language, prompt, temperature, model alias, and response format without writing a client first.

Microphone recording and drag-and-drop upload
JSON, verbose JSON, and plain-text responses
Segment timing, request metadata, and copyable cURL

Open the Transcription Studio

Add audio

WAV, MP3, M4A, FLAC, OGG, or WebM—up to 25 MB.

Choose a route

Swedish and English use independent, long-lived workers.

Use the transcript

Copy text, inspect segments, or download the normalized response.

Quick start

One multipart request.

The public API follows the familiar OpenAI transcription shape while requiring an explicit language.

Full API reference Try it visually

bash

curl http://localhost:8080/v1/audio/transcriptions \
  -F file=@recording.wav \
  -F language=sv \
  -F response_format=verbose_json