Swiss German speech-to-text with state-of-the-art quality
Our API transcribes Swiss German, German, French, Italian, English and many more languages – operated in Switzerland, compliant with the nFADP and GDPR, with speaker diarization, automatic language detection and accurate timestamps.
State-of-the-art quality for every dialect region
We have extensive, high-quality training data of our own for Swiss German, covering every Swiss German dialect region: Basel, Bern, Grisons, Eastern Switzerland, Valais, Central Switzerland and Zurich. We train our speech-to-text models on this data and reach state-of-the-art quality in every dialect region.
Our asynchronous REST API. Integrated in an hour.
Upload audio, receive a request ID, fetch the result – segmented, with speaker diarization and accurate word-level timestamps.
You can try the API directly in your browser via the API documentation. For recordings of 5 minutes or longer, we recommend testing programmatically, e.g. with curl. When you test via the OpenAPI documentation, it shows the curl commands it ran.
Available models:
- - whisper-ch: model specialised in Swiss German
- - whisper-orig: general-purpose standard model, recommended for audio without Swiss German
- - Model parameter not set: whisper-ch if language = de-CH, otherwise whisper-orig
{
"state": 0,
"message": "Request successfully completed",
"response": {
"content_format": "json",
"content": [
{
"start": "00:00:00.4520000",
"end": "00:00:06.3820000",
"transcript": "Wie viele Kilometer hast du mit deinem Auto schon gemacht inzwischen?",
"speaker_id": 1,
"words": [
{
"start": "00:00:00.4520000",
"end": "00:00:00.6720000",
"transcript": "Wie",
"speaker_id": 1,
"score": 0.545
},
{
"start": "00:00:00.6920000",
"end": "00:00:00.9130000",
"transcript": "viele",
"speaker_id": 1,
"score": 0.481
},
{
"start": "00:00:00.9530000",
"end": "00:00:01.7340000",
"transcript": "Kilometer",
"speaker_id": 1,
"score": 0.787
},
{
"start": "00:00:01.7740000",
"end": "00:00:01.9340000",
"transcript": "hast",
"speaker_id": 1,
"score": 0.199
},
{
"start": "00:00:02.0550000",
"end": "00:00:02.1550000",
"transcript": "du",
"speaker_id": 1,
"score": 0.522
},
{
"start": "00:00:02.1750000",
"end": "00:00:02.4150000",
"transcript": "mit",
"speaker_id": 1,
"score": 0.631
},
{
"start": "00:00:02.4550000",
"end": "00:00:02.6560000",
"transcript": "deinem",
"speaker_id": 1,
"score": 0.356
},
{
"start": "00:00:02.6760000",
"end": "00:00:03.0560000",
"transcript": "Auto",
"speaker_id": 1,
"score": 0.826
},
{
"start": "00:00:03.0760000",
"end": "00:00:03.2370000",
"transcript": "schon",
"speaker_id": 1,
"score": 0.521
},
{
"start": "00:00:03.2570000",
"end": "00:00:03.6970000",
"transcript": "gemacht",
"speaker_id": 1,
"score": 0.603
},
{
"start": "00:00:03.7370000",
"end": "00:00:06.3820000",
"transcript": "inzwischen?",
"speaker_id": 1,
"score": 0.75
}
]
},
{
"start": "00:00:06.4220000",
"end": "00:00:09.7280000",
"transcript": "Aktuell hat er noch keine 30000 Kilometer auf dem Tacho.",
"speaker_id": 0,
"words": [
{
"start": "00:00:06.4220000",
"end": "00:00:06.9630000",
"transcript": "Aktuell",
"speaker_id": 0,
"score": 0.734
},
{
"start": "00:00:06.9830000",
"end": "00:00:07.1030000",
"transcript": "hat",
"speaker_id": 0,
"score": 0.65
},
{
"start": "00:00:07.1430000",
"end": "00:00:07.2240000",
"transcript": "er",
"speaker_id": 0,
"score": 0.775
},
{
"start": "00:00:07.2640000",
"end": "00:00:07.3840000",
"transcript": "noch",
"speaker_id": 0,
"score": 0.654
},
{
"start": "00:00:07.4240000",
"end": "00:00:07.7040000",
"transcript": "keine",
"speaker_id": 0,
"score": 0.692
},
{
"start": "00:00:07.7850000",
"end": "00:00:08.5060000",
"transcript": "30000",
"speaker_id": 0,
"score": 0.757
},
{
"start": "00:00:08.5460000",
"end": "00:00:09.0070000",
"transcript": "Kilometer",
"speaker_id": 0,
"score": 0.678
},
{
"start": "00:00:09.0670000",
"end": "00:00:09.1470000",
"transcript": "auf",
"speaker_id": 0,
"score": 0.363
},
{
"start": "00:00:09.1670000",
"end": "00:00:09.3470000",
"transcript": "dem",
"speaker_id": 0,
"score": 0.385
},
{
"start": "00:00:09.3870000",
"end": "00:00:09.7280000",
"transcript": "Tacho.",
"speaker_id": 0,
"score": 0.323
}
]
}
],
"language": "de-CH",
"speaker_count": 2,
"duration": "00:00:10.4080000",
"measured_duration": "00:00:09.2360000",
"input_size": 333056,
"message": ""
}
}
Looking for an easy-to-use web interface and editor?
Want the same speech-to-text quality with an easy-to-use web interface and an editor for polishing transcripts by hand? Then Stenoris is the perfect fit for you.
The solution for parliaments and commissions
Together with our partner it-processing AG we offer Verbalix, a professional transcription solution for political committees like parliaments and commissions.
Save valuable time taking meeting minutes with our built-in speech-to-text system, which, of course, also supports Swiss German at state-of-the-art quality. Speech-to-text in Verbalix runs on a model optimized specifically for political committees.
Our customers
«The quality of transcription from Swiss German to Standard German is, in our view, at a very high level and truly impressive.»
«As one of the first customers of Ateleris for our order automation service, we have been nothing but happy! It is fast, accurate, and affordable, with a team that genuinely cares about their customers.»
At the forefront of Swiss German speech-to-text since 2019
Ateleris experts Michel Plüss and Julia Hartmann played a leading role in seven publications on Swiss German speech-to-text and were first author on six of them.
Research and engineering in one team
At the forefront of research
Our team publishes on Swiss German speech-to-text and works closely with the Institute for Data Science at FHNW. New methods reach the product quickly.
MLOps, not one-off experiments
Versioned data, reproducible training, automatic evaluation on fixed test sets. Every release is measurably better, or it does not ship.
Current models
We continuously track new architectures and keep training them on our data. Improvements reach you automatically through the same API, with no migration effort.
Space-grade quality
We also build software for ESA missions. The same standards for code reviews, testing and documentation apply to our transcription service.
Test it with your own recordings.
We're happy to send you a test API key and answer any questions.