← All Piper voices

`de_DE-thorsten-medium`

de_DE-thorsten-medium is a German (Deutsch) voice for the Piper engine, and Quick TTS runs it free in your browser with nothing uploaded. This page is the honest version: what tier it is, what sits beneath it when your device cannot load it, what the alternative on German is, and where it falls short.

What this voice actually is

Fact	Value
Voice ID	`de_DE-thorsten-medium`
Engine	Piper, running as WebAssembly in your browser
Language	German (Deutsch)
Quality tier	`medium`, the tier Quick TTS treats as the default everywhere, and the best quality-per-megabyte Piper offers
Preselected?	Yes. This is the voice Quick TTS starts on when you choose Piper on German.
Needs a GPU?	No. Piper is the engine that runs without one.
Step-down tier	None. This voice is single-rung, so a device that cannot load it falls back to the browser engine rather than stepping down to a smaller model.
Also available in HD?	Yes. Supertonic HD speaks German at 44.1kHz, on WebGPU only.
Integrity	SHA-256 pinned. Weights are rejected unless their hash matches the pin, on every host we fetch from.

What is worth knowing about it

German is single-rung here. de_DE-thorsten-low and de_DE-eva_k-x_low both exist upstream and both throw OrtRun ERROR_CODE 2 at inference in the pinned release, so German has no working step-down: a device that cannot allocate this model's session falls back rather than down-tiering. Supertonic HD also speaks German, so there is still a second neural option; it just needs a GPU instead of spare memory.

It is the only selectable German Piper voice

German ships exactly one Piper voice in Quick TTS. That is an upstream constraint rather than an editorial one: the voices that exist in the catalog but are not offered here either fail at inference in the pinned release or do not exist at all. We would rather hide a voice than offer one that fails on the play click.

What happens when your device cannot load it

Piper's real failure mode is memory, not speed. The model has to allocate an ONNX session, and on weak or GPU-less hardware that allocation is the thing that fails. Quick TTS handles it in two layers, and both are automatic:

Before the download. A device reporting 4GB of memory or less starts on the smallest available tier instead of the default, so it never pulls the large model only to fail on it. Below 2GB, Piper is hidden entirely rather than offered and then failing.
After a failure. German has no smaller working model, so there is nothing to step down to. The remainder of the text is handed back to the browser's built-in speech engine instead, and Supertonic HD is the other neural option if the machine has a working GPU.

How to use it

Open quick-tts.com/de/, switch the engine dropdown to Piper, and choose de_DE-thorsten-medium from the voice list. The neural voice list follows the page language, so German voices appear on the German page rather than on the English root. Speed, volume and voice can all be changed mid-read and the current passage replays under the new settings without starting over. Switching the engine itself stops playback, so you press play once to start the new one. There is no mid-read handover of the remaining text between engines.

The first play click downloads the model once and caches it; afterwards it starts immediately and keeps working offline.

Nothing you paste is uploaded

Piper runs as WebAssembly inside your own browser tab. The model is downloaded once and cached; after that, the text you paste is turned into audio on your machine and never sent anywhere. That is an architectural property rather than a policy promise: the test is that generation keeps working with your network disconnected once the model has loaded.

Common questions

What is de_DE-thorsten-medium?

It is a German neural text-to-speech voice for the Piper engine, at the medium quality tier. Quick TTS runs it as WebAssembly inside your browser, so it needs no GPU and no account, and the text you paste is never uploaded.

Is de_DE-thorsten-medium free?

Yes. There is no account, no character limit and no watermark. The model downloads once to your browser and then runs on your own device, so there is no per-character server cost to meter.

Do I need a GPU to use de_DE-thorsten-medium?

No. Piper runs on WebAssembly and works on machines with no usable GPU at all, which is why it is the most-selected engine here. The GPU-only alternative on German is Supertonic HD.

How do I select de_DE-thorsten-medium in Quick TTS?

Open quick-tts.com/de/, switch the engine dropdown to Piper, and pick it from the voice list. The neural voice list follows the page language, so you need the German page to see German voices.

Can I download the audio from de_DE-thorsten-medium?

Yes, as WAV or MP3, generated on your device. Anything past roughly 20,000 characters is packaged as a ZIP of audio parts so that memory stays bounded on long documents.

de_DE-thorsten-medium