🌸 Suzune Mini (40M On-Device Causal Flow TTS)

Suzune Mini is an ultra-compact, 40-million parameter on-device Text-to-Speech (TTS) foundation model engineered by Sorika Labs.

Powered by causal flow-matching and neural latent acoustic decoding, Suzune Mini delivers first-audio latency under 10ms directly on laptop CPUs, mobile devices, edge compute, and Apple Silicon, with zero network round-trip.

🏛️ Base Model & Lineage

Lineage & Attribution: Suzune S1 Mini is developed and released by Sorika Labs. Its underlying causal flow-matching acoustic architecture is adapted and fine-tuned from the open-source gradium/phonon-31m-en foundation (released under Apache 2.0).

Sorika Labs Enhancements & Customizations:

1. Calibrated High-Fidelity Voice Manifolds: Built-in voice profiles including Zoey, Harper, Easton, Juniper, Austin, and Aria.
2. Causal Acoustic Optimization: Streamlined latent flow generation for sub-10ms real-time conversational agents.
3. Standalone suzune-mini SDK: Clean, high-level Python API with zero complex boilerplate.

🚀 Quick Install

pip install suzune-mini

🎙️ Python Quickstart

Synthesize speech in 3 lines of code:

import suzune_mini

# Load Suzune Mini engine (default voice: 'Zoey')
tts = suzune_mini.load(voice="Zoey")

# Generate speech
audio, sr = tts.speak(
    "Hello! Suzune Mini is running 100% on device with sub-10ms latency.",
    output="speech.wav"
)

🎭 Available Built-in Voices

💻 Terminal Command Line (CLI)

suzune-mini "Hello from Sorika Labs on-device AI!" --voice Zoey --output output.wav

📜 License & Citation

Licensed under the Apache 2.0 License.

@misc{sorika2026suzunemini,
  author = {Sorika Labs},
  title = {Suzune S1 Mini: Ultra-Lightweight On-Device Flow-Matching Speech Foundation Model},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {https://huggingface.co/sorika-labs/suzune-s1-mini}
}