Accept the BreezeBlue model license
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
By requesting access, you confirm that you have read and accept the BreezeBlue Research and Non-Commercial License Agreement and will use this derivative only for research or non-commercial purposes.
Log in or Sign Up to review the conditions and access this model content.
Sirocco MLX BF16
Derived from Breeze TTS 2 by BreezeBlue and licensed for research and non-commercial use only.
Sirocco MLX BF16 is a BF16 Apple Silicon MLX conversion, pinned to upstream model revision
c1c8ca18b70b30822735633991d9ebf4898e47d4. It is intended for the
standalone mlx-breeze-tts2
runtime on Apple Silicon.
Capabilities
- English and Chinese voice design
- Zero-shot voice cloning with exact reference transcript
- Reference-guided voice direction
- English and Chinese vocal events
- Non-streaming and incremental streaming generation
- Seed, CFG, temperature, top-p, top-k, repetition penalty, and token controls
- Python, CLI, and upstream-compatible HTTP interfaces
Conversion
Precision: BF16.
No affine quantization. This is the behavior-alignment reference artifact.
The conversion records strict key/shape audits, upstream provenance, and weight hashes in the included configuration and audit files.
Local M3 Max evaluation
The release passed a 23-case real-device matrix, Whisper content checks, SpeechBrain ECAPA identity checks, waveform and streaming integrity checks, revision-bound PyTorch parity, real HTTP probing, and human review of all eight English/Chinese event types.
- Corpus CER:
0.01042 - Minimum clone cosine / segment P10:
0.5992 / 0.5940 - Steady-state RTF:
3.777 - Streaming TTFA:
3.603 s - Peak memory:
12.03 GB - Maximum reference leakage:
0.1695 - Manual event audibility:
8/8 pass - Final independent release verifier:
pass
Full evidence is in the project release bundle.
Usage
Install and authenticate for this gated repository:
pip install 'git+https://github.com/vanch007/mlx-breeze-tts2.git'
hf auth login
Generate speech:
mlx-breeze-tts2 generate \
--model vanch007/Sirocco-MLX-BF16 \
--text "(sigh) Welcome aboard. Your journey begins now." \
--instruction "A warm, thoughtful voice with a calm delivery." \
--cfg-scale 4 --seed 7 --output output.wav
See the project README for cloning, direction, streaming, HTTP serving, and conversion examples.
License and responsible use
The model weights, tokenizer/codec weights, derivative parameters, and self-hosted outputs are governed by the included BreezeBlue Research and Non-Commercial License Agreement. Commercial use is not permitted without a separate written license from BreezeBlue.
You must obtain all necessary rights and consent for reference audio and voices. Unauthorized impersonation, deceptive synthetic audio, fraud, harmful use, and unlawful biometric use are prohibited. This repository is an independent community conversion and is not affiliated with or endorsed by BreezeBlue.
- Downloads last month
- -
Quantized
Model tree for vanch007/Sirocco-MLX-BF16
Base model
BreezeBlue/Breeze-TTS-2