MIO-TTS-V1

MIO TTS V1

MIO TTS V1 is an experimental German Piper TTS voice created and trained by Maurice Hartmann.

This release is the first publicly usable milestone of an ongoing scratch-training run.

Current training state

• Architecture: Piper / VITS
• Quality preset: High
• Language: German
• Training method: Scratch training
• Batch size: 24
• Maximum phoneme IDs: 500
• Checkpoint: epoch=489-step=278320
• Completed epochs: 490
• Export format: ONNX
• Sample rate: 22050 Hz
• Status: Work in progress

Training is continuing. Later releases are expected to improve pronunciation, prosody, stability, and long-sentence handling.

Intended use

• fast local text-to-speech
• assistant and agent communication
• offline voice output
• experimentation with Piper
• German-language TTS applications

This version is already considered practically usable for fast communication, but it is not the final trained model.

Files

• mio_tts_v1.onnx
• mio_tts_v1.onnx.json
• mio_tts_v1_demo.wav

Example usage

echo "Guten Morgen. Dies ist MIO TTS Version eins." | piper --model mio_tts_v1.onnx --output_file output.wav

Training dataset

MIO TTS V1 is based directly on the publicly available synthetic German training dataset:

Muckylixx/german-sohee-synthetic-tts-24k

This dataset was created and published by Maurice Hartmann specifically for the development of German TTS voices.

It contains approximately 19.5 hours of German synthetic speech and represents the complete training foundation used for this first MIO TTS release.

By publishing both the dataset and the resulting model, the project aims to make the development process transparent and reproducible.

Future MIO TTS releases will continue to build on this training material and may include improved preprocessing, additional data, and later training checkpoints.

Community and feedback

MIO TTS is an active community-oriented project and is still under development.

Feedback, test results, pronunciation reports, technical suggestions, and ideas for future improvements are very welcome.

Especially useful feedback includes pronunciation errors, unusual pauses, long-sentence problems, Piper integration experiences, comparisons with other German voices, and suggestions for later releases.

Comments, discussions, and constructive criticism are explicitly encouraged. Community feedback may directly influence later versions of MIO TTS.

Please feel welcome to use the Hugging Face Community section of this model repository.

Limitations

This is an early training milestone. Possible remaining limitations include inconsistent prosody, pronunciation errors, unusual pauses, instability on very long text, and artifacts in uncommon words or abbreviations.

Author

Maurice Hartmann
Hugging Face: Muckylixx

License

Custom terms are included in the LICENSE file.

Copyright © 2026 Maurice Hartmann.