OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Ming Uniaudio

Ming Uniaudio is an open-source Python speech-model project that combines speech understanding and generation through a shared continuous audio tokenizer, and provides a dedicated model for instruction-guided free-form speech editing. Its GitHub repository includes model and benchmark downloads, example usage, and supervised fine-tuning recipes for speech generation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Ming Uniaudio — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Ming Uniaudio when you want to work directly with a unified speech model and can manage its code, model weights and local runtime, rather than use a packaged audio-editing app.

Who it is for

  • Speech-AI researchers exploring a shared model for understanding and generation.
  • Developers building speech-generation or instruction-guided audio-editing systems.
  • Teams seeking a repository example or fine-tuning recipes for speech generation.

Documented uses

  • Use speech input for understanding tasks with a unified speech model.
  • Generate speech using the model framework and its speech-generation fine-tuning recipes.
  • Edit speech through natural-language instructions without specifying edit regions.
  • Explore or evaluate free-form speech editing using the linked evaluation set.
Before you choose

Requirements & limits

The repository is public, identifies the project as MIT-licensed, and links to model downloads on Hugging Face and ModelScope.

  • The documented example requires both source code and separately downloaded model weights, linked into the repository’s expected directory.
  • The README describes setup through pip requirements or Docker and provides a Python example script.
  • Reported example hardware is NVIDIA H800-80GB or H20-96G with CUDA 12.4.
  • The README marks vLLM inference support as pending.
  • Speech-generation supervised fine-tuning is documented for full-parameter and LoRA approaches.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare and run the repository example

Set up the documented source-and-weights layout, then invoke the provided example script.

Before you start

  • Ming-UniAudio source repository
  • Downloaded Ming-UniAudio model weights
  • Python environment with dependencies from requirements.txt, or a Docker environment
  • NVIDIA H800-80GB or H20-96G with CUDA 12.4 are the hardware conditions stated for the authors’ examples

Steps

  1. Download the source repository and change into its directory.

  2. Download the Ming-UniAudio model weights using the repository’s model-download instructions.

  3. Create an inclusionAI directory and a symbolic link named Ming-UniAudio-16B-A3B pointing to the downloaded weights.

  4. Run python3 cookbooks/test.py from the repository directory.

Expected outcome

The repository provides a Python example invocation for the model; the README points to demo.ipynb for detailed usage.

Guide limitations

  • The README reports example hardware conditions but does not establish compatibility with other hardware or environments.
  • This recipe describes the published example steps and is not a verified execution test.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
inclusionAI/Ming-UniAudio · Official homepage title · retained in its source language
How we review sourcesSuggest a correction