MOSS-TTS

MOSS-TTS

Generate local multilingual speech with the open-source MOSS-TTS family from MOSI.AI and OpenMOSS. This skill focuses on MOSS-TTS-Nano first because it is lightweight, CPU friendly, and practical for local RAPR workflows, while documenting upgrade paths for the full MOSS-TTS family.

What you can do

• Local TTS: Convert text into speech without a hosted API.
• CPU-friendly demos: Use MOSS-TTS-Nano ONNX for lightweight local inference.
• Voice cloning: Generate speech from a short local reference audio clip when you have rights or consent.
• Multilingual narration: Work across English, Chinese, and other supported languages.
• Dialogue workflows: Use MOSS-TTSD for expressive multi-speaker spoken dialogue when you need the heavier model path.
• Audio generation: Explore the broader MOSS-TTS family for voice design and sound effect generation.

Prerequisites

A clean Python environment is recommended. The upstream project recommends Conda for the easiest setup:

conda create -n moss-tts-nano python=3.12 -y
conda activate moss-tts-nano
git clone https://github.com/OpenMOSS/MOSS-TTS-Nano.git
cd MOSS-TTS-Nano
pip install -r requirements.txt
pip install -e .

For local CPU inference, start with the ONNX entrypoints:

python infer_onnx.py --prompt-audio-path assets/audio/zh_1.wav --text "Welcome to MOSS-TTS-Nano."
python app_onnx.py

Then open http://127.0.0.1:18083 for the local web demo.

Example Requests

• "Generate a narration WAV from this paragraph using MOSS-TTS-Nano."
• "Start the local MOSS-TTS web demo."
• "Use this reference audio to clone a voice for a consented demo."
• "Create separate audio files for each section of this article."
• "Compare the Nano CPU workflow with the heavier MOSS-TTS dialogue model."

Use voice cloning only with consent and appropriate rights. Do not use this skill for impersonation, fraud, or bypassing identity checks.