Kikiri German — Martin Harbecke
German single-speaker TTS model fine-tuned on the Martin Harbecke voice using StyleTTS2 Stage 2.
Built on top of kikiri-german-base-51speakers-synthetic.
Demo
StyleTTS 2 — Abstract (German)
Model Details
| Property | Value |
|---|---|
| Architecture | StyleTTS2 Stage 2 (Kokoro-compatible) |
| Language | German (de) |
| Speaker | Martin Harbecke (synthetic) |
| Training samples | 627 |
| Training epochs | 10 |
| WavLM discriminator | active from epoch 4 |
| Mixed precision | bf16 |
| Sample rate | 24 kHz |
Usage
# Uses the kokoro library as underlying framework
from kokoro import KModel, KPipeline
import torch
kmodel = KModel(repo_id="hexgrad/Kokoro-82M", config="config.json", model="kikiri_german_martin_ep10.pth").to("cuda").eval()
pipeline = KPipeline(lang_code="de", model=kmodel)
voice = torch.load("voices/martin.pt", map_location="cpu")
text = "Guten Tag, wie geht es Ihnen?"
for _, _, audio in pipeline(text, voice=voice):
# audio is a numpy array at 24kHz
pass
Files
| File | Description |
|---|---|
kikiri_german_martin_ep10.pth |
Kokoro-compatible model weights (Stage 2, epoch 10) |
voices/martin.pt |
Martin speaker voicepack (510, 1, 256) |
audio/styletts2_demo.wav |
Demo: StyleTTS 2 abstract read by Martin |
License
Apache 2.0