Sonus

Sonus

A massively multilingual zero-shot text-to-speech synthesis system

Overview

Sonus is an advanced multilingual zero-shot text-to-speech synthesis system supporting over 600 languages. Built on a novel architecture, it delivers high-quality speech generation with superior inference speed, supporting voice cloning and voice design capabilities.

Key Features

• 600+ Languages Supported: Broad language coverage for zero-shot TTS
• Voice Cloning: High-quality voice cloning from short reference audio
• Voice Design: Control voices via speaker attributes (gender, age, pitch, accent, etc.)
• Fine-grained Control: Support for non-verbal symbols and pronunciation correction
• Fast Inference: Optimized for real-time and batch processing

Installation

pip install torch torchaudio
pip install transformers

Quick Start

from transformers import AutoModel, AutoTokenizer
import torch

model = AutoModel.from_pretrained("comethrusws/sonus", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("comethrusws/sonus", trust_remote_code=True)

# Load to device
model = model.to("cuda")

# Generate speech
text = "Hello, this is a test of voice synthesis."

Model Specifications

• Architecture: Diffusion language model-style
• Parameters: 0.6B
• Sampling Rate: 24 kHz
• Languages: 600+

License

This project is available under a custom license.

• Non-commercial use: Free for personal projects, research, and educational purposes
• Commercial use: Requires explicit permission. Contact [email protected] for licensing inquiries

See LICENSE file for full terms.

Disclaimer

Users are prohibited from using this model for unauthorized voice cloning, impersonation, fraud, or any illegal activities. Ensure compliance with applicable laws and ethical standards.