kikiri-tts/hui-german-51speakers-synthetic

---
language:
- de
license: other
task_categories:
- text-to-speech
tags:
- tts
- german
- multispeaker
- kikiri-tts
- synthetic
pretty_name: HUI German 51 Speakers Synthetic
size_categories:
- 10K<n<100K
---

# HUI German 51 Speakers Synthetic

Multispeaker German TTS dataset for Kikiri TTS Stage 1 training.

## Dataset

- **51 speakers** from the [HUI-Audio-Corpus-German](https://github.com/iisys-hof/HUI-Audio-Corpus-German)
- **~102h** synthetic audio, ~22,700 clips
- Audio synthesized with **Qwen3-TTS** from original HUI transcript texts
- Clips segmented to max ~15s, 22kHz mono WAV

## Format

```
wavs/<speaker_id>/<filename>.wav
train_list.txt
val_list.txt
```

`train_list.txt` format (multispeaker): `<speaker_idx>|<wav_path>|<transcript>`

## Usage

Designed for Kikiri TTS Stage 1 multispeaker training.
Training workflow based on [semidark/hokuspokus-qwen3-tts-hybrid](https://huggingface.co/datasets/semidark/hokuspokus-qwen3-tts-hybrid).

## Credits

- Speaker recordings basis: [HUI-Audio-Corpus-German](https://github.com/iisys-hof/HUI-Audio-Corpus-German) (Florian Lux et al.)
- Training workflow & data pipeline: [@semidark](https://github.com/semidark)
- Audio synthesis: Qwen3-TTS

## License

No formal license declared. The HUI-Audio-Corpus-German is based on LibriVox recordings (Public Domain in the USA; legal status varies by country). The HUI team (Hochschule Hof) requests attribution in the spirit of CC-BY-SA 4.0 but does not legally enforce it.

Synthetic audio in this dataset was generated from HUI transcript texts using Qwen3-TTS; no original HUI audio is included.