Thorsten-Voice/tv-orpheus-v1

Thorsten-Voice – Orpheus TTS v1 (German)

Overview

Thorsten-Voice/tv-orpheus-v1 is a German text-to-speech (TTS) model trained with Orpheus TTS.
It is based on the German Orpheus research checkpoint canopylabs/3b-de-ft-research_release and fine-tuned with approximately 12,000 recordings from the Thorsten-Voice Dataset 2022.10.

The goal of this model is to provide a natural, calm, and intelligible German male voice that is freely usable for research, education, and creative projects.

Training date / codebase reference: December 2025

---

Training Data

• Dataset: Thorsten-Voice Dataset 2022.10
• Number of recordings: ~12,000
• Sample rate: 24 kHz
• Speaker: Thorsten (single speaker)
• License of recordings: CC0 (Public Domain)
• DOI: https://doi.org/10.5281/zenodo.7265581

The speaker explicitly releases the voice recordings and the underlying voice under CC0.

---

Model Lineage

• Base model: canopylabs/3b-de-ft-research_release
• TTS framework: Orpheus TTS
• Fine-tuning: Thorsten-Voice (Germany)
• Architecture: Orpheus-style LLM + SNAC codec

This model is a derivative work of the Orpheus base model.

---

License Clarification

• Model weights:
  Licensed under Apache-2.0, inherited from the Orpheus base model.

• Training data & voice:
  Released under CC0 (Public Domain) by the speaker.

No additional usage restrictions are imposed by the author.

---

Usage Example

python inference_hf.py \
  --model_path Thorsten-Voice/tv-orpheus-v1 \
  --text "Guten Morgen. Ich hoffe, es geht Ihnen gut." \
  --outfile output.wav

The inference script (inference_hf.py) is included in this repository (see files).

Acknowledgements

This model would not exist without the excellent work on Orpheus TTS.

• Orpheus TTS GitHub: https://github.com/canopylabs/orpheus-tts
• Base model by: Canopy Labs

Thorsten-Voice Project

• Thorsten-Voice is an open voice project providing high-quality German speech data and models under open licenses.
• Project page: https://www.Thorsten-Voice.de