eigenhand/Articulatio-DE-MLX

Articulatio-DE MLX (4 bit)

4-bit MLX conversion of Articulatio-DE, German text-to-speech with zero-shot voice cloning, for Apple Silicon. Base weights of Breeze TTS 2 with the Articulatio-DE adapter merged in, converted with mlx_audio.convert --quantize --q-bits 4 (mlx-audio 0.5.6).

Derived from Breeze TTS 2 by BreezeBlue and licensed for research and non-commercial use only.

Articulatio-DE is an independent community project. It is not affiliated with, endorsed by, or an official release of BreezeBlue or RESONIA, INC.

Sample

„Guten Morgen! Heute ist Donnerstag, der fünfundzwanzigste September. Draußen sind es achtzehn Grad, am Nachmittag zieht von Westen ein Gewitter auf. Vergiss also den Regenschirm nicht, wenn du später noch zum Bahnhof fährst.“

Synthetic speech, generated with Articulatio-DE (GGUF Q8_0, one sentence per piece, speed 1.25 with the pitch preserved, 150 ms pause between sentences) in a voice the model created itself; it belongs to no real person. The clip is an output of the model and falls under the same license.

On 130 held-out test sentences (cloned voices, Whisper large-v3) it reaches 4.6 % WER in German and 3.1 % in English; the real recordings of the same sentences give 4.4 % and 2.6 %. Details in the main model card.

Usage

With articulatio-mlx, about 1.5x faster than realtime on an M3 Pro (RTF 0.69, first audio after ~0.4 s when streaming):

hf download eigenhand/Articulatio-DE-MLX --local-dir articulatio-de-mlx
articulatio-mlx articulatio-de-mlx --text "Guten Morgen, heute ist ein schöner Tag." \
  --ref-audio reference.wav --ref-text "Exakter Wortlaut der Sprachprobe." --output out.wav

It also loads in mlx-audio 0.5.6 directly, about three times slower. Pass one sentence at a time: in one piece, passages of more than about 20 s derail. Write out digits and symbols first, for example with the German text frontend. Vocal event tags such as (laugh) only rarely work in German with this model. Everything else — instructions, cfg_scale, limitations — is in the main model card.

License

Distributed under the BreezeBlue Research and Non-Commercial License Agreement, the license of Breeze TTS 2; see also NOTICE. Clone only voices whose speakers have explicitly agreed, and use the model and its outputs only non-commercially.

Breeze TTS 2 is licensed under the BreezeBlue Research and Non-Commercial License Agreement. Copyright (c) 2026 RESONIA, INC. All Rights Reserved.