# German audio voices

These are generated speech recordings, not recordings of a person reading the course.

- **Thorsten**, Piper `de_DE-thorsten-high`: [Thorsten Voice](https://github.com/thorstenMueller/Thorsten-Voice), CC0 voice data; Piper model distributed under MIT terms.
- **Kerstin**, Piper `de_DE-kerstin-low`: [voice dataset](https://github.com/rhasspy/dataset-voice-kerstin), CC0. The [model card](https://huggingface.co/rhasspy/piper-voices/blob/main/de/de_DE/kerstin/low/MODEL_CARD) specifies German, one speaker and a 16 kHz sample rate.

Models are provided through the [sherpa-onnx TTS release](https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models) and run locally using sherpa-onnx. The six A1 mock dialogues alternate Thorsten and Kerstin with short pauses between turns. Other course lines use Thorsten. Existing recordings are retained when their text is unchanged.

Generated speech is a practice aid. Use the linked official Goethe recordings for further listening rehearsals. These generated clips have not received independent human listening approval.