[FEATURE] Support MOSS-TTS-Nano

Describe the feature

Add support for the ONNX version of MOSS-TTS-Nano in sherpa-onnx, including model loading, text-to-speech inference, streaming generation, and cross-platform deployment support.

Target model:

• MOSS-TTS-Nano: https://github.com/OpenMOSS/MOSS-TTS-Nano
• ONNX weights: https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Nano-100M-ONNX

MOSS-TTS-Nano is a lightweight multilingual TTS model designed for realtime speech generation on CPU-only devices. Supporting this model in sherpa-onnx would allow developers to deploy high-quality multilingual speech synthesis directly on mobile, desktop, embedded, and edge devices using the existing sherpa-onnx ecosystem.

Motivation / Use Case

MOSS-TTS-Nano provides several capabilities that make it highly suitable for sherpa-onnx users:

• Small model size (~0.1B parameters)
• Realtime CPU inference
• Multilingual speech synthesis
• Streaming generation support
• Voice cloning capability
• ONNX-native deployment workflow
• Cross-platform compatibility

Many developers are already using sherpa-onnx as a unified runtime for speech applications. Adding support for MOSS-TTS-Nano would enable:

• Offline TTS applications
• AI voice assistants
• Audiobook readers
• Accessibility tools
• Mobile and edge-device speech synthesis
• Flutter, Android, iOS, macOS, Windows, Linux, and embedded deployments

This would also strengthen sherpa-onnx's TTS ecosystem by providing an additional modern open-source multilingual model alongside existing supported speech models.

Additional context

According to the official project, MOSS-TTS-Nano is an open-source multilingual TTS model from the OpenMOSS team focused on lightweight deployment and realtime inference. The ONNX version is specifically optimized for CPU execution and can run efficiently without PyTorch dependencies. (GitHub)

Relevant resources:

• MOSS-TTS-Nano repository: https://github.com/OpenMOSS/MOSS-TTS-Nano
• MOSS-TTS repository: https://github.com/OpenMOSS/MOSS-TTS
• Technical report: https://arxiv.org/abs/2603.18090

Potential implementation scope:

• ONNX model loading
• Streaming TTS inference
• Multilingual synthesis
• Speaker reference / voice cloning support (if applicable)
• C++ API support
• Python API support
• Dart / Flutter bindings support
• Example applications and documentation