MOSS-SoundEffect without CUDA (Mac and AMD): audio generator — what it does and what you need to run it

MOSS-SoundEffect without CUDA is a desktop application for generating sound effects from a text description. This build targets Mac computers with Apple Silicon chips and Windows machines with AMD GPUs that lack CUDA support. It runs locally, without a browser and without access keys to external services.

What it does

At its core is the MOSS-SoundEffect-v2.0 diffusion model for text-to-audio conversion. As described by the author, the application includes a desktop interface, a command-line interface, performance measurement tools, and an MCP server. The authors note that no API keys or browser are required.

The model, according to the developer, requires roughly 8–10 GB of system RAM or VRAM. On Apple Silicon, device memory is used, so the authors do not provide an exact split between RAM and VRAM.

What you need to run it

Supported platforms as described by the author are AMD ROCm and Apple Silicon. The code is written in Python, with the tested interpreter version being 3.12. The author notes that the build was tested on an Apple M4 Pro with macOS 26 in bfloat16 mode.

The code is open source and distributed under the Apache-2.0 license. The repository is hosted at github.com/VladimirTalyzin/MOSS-SoundEffect_v2.0_MPS_ROCm. The latest code change is dated July 22, 2026.

Who it is for

This build is intended for those working on a Mac with Apple Silicon or a Windows machine with an AMD GPU who want to generate sound effects locally. The lack of a CUDA dependency makes it an option for hardware fleets without NVIDIA GPUs.

Users with NVIDIA GPUs do not need this build — the main CUDA version exists for them. Memory requirements should also be considered: if the device has less than 8–10 GB of available memory, running the model may be difficult or impossible.

The project is young: the repository was created on July 22, 2026, and the latest change was made the same day. This is worth keeping in mind when assessing stability and the amount of user experience.