RepoPicks Radar

Updated every day around 23:30 UTC

Guides

AudioLDM 2

haoheliu/AudioLDM2Reviewed by AI

Checked by RepoPicks Radar on

What is AudioLDM 2?

A model that turns a written description into audio: sound effects, short music, or speech from a transcription plus a description of the speaker. It also upgrades the quality of an existing recording. You can try it free on the authors' Hugging Face pages, or install it with Python and run either a local web page or a one-line command. Seven different model files are available, including a 48 kHz high-fidelity one. It is research code from a published paper, last changed in 2024.

AudioLDM 2 is free and can replace ElevenLabs, Suno, and Epidemic Sound.

How easy is it to install AudioLDM 2?

A · download and open · Web

  • Windows C
  • macOS C
  • Linux C
  • Web A

The easiest way in is the web, an A: the README links two free Hugging Face Spaces run by the author where you type a description and listen, with nothing installed. Installing it is a C on all three systems: create a Python 3.8 environment with conda, `pip3 install git+https://github.com/haoheliu/AudioLDM2.git`, then either `python3 app.py` for a local web page or `audioldm2 -t "..."` in the terminal. It can use an NVIDIA card (cuda), an Apple Silicon Mac (mps) or plain cpu, and text-to-speech also needs espeak installed.

AudioLDM 2 on GitHub

Stars
2,644on Nov 10, 2026
Downloads
—all releases
Growth
+1stars in 6 days

Stars over time

2,6442,6452,646 Nov 6Nov 7Nov 8Nov 9Nov 10 2,6442,6452,646 Nov 6Nov 8Nov 10

Facts

Runs on
Windows, macOS, Linux, Web
You need to install
Python, NVIDIA graphics card
To build from source
Python, NVIDIA graphics card
Disk space
950 MB–4.3 GB · estimate
Accounts or keys
None
Cost
Free · No price, but the LICENSE file is Creative Commons Attribution-NonCommercial-ShareAlike 4.0, which does not allow commercial use — while setup.py declares MIT.
License
NOASSERTION (Other)
Last change on GitHub
Sep 29, 2024

AudioLDM 2 risk signals (8)

  • AttentionDependencies with known vulnerabilities

    Pinned dependencies with known vulnerabilities (OSV): transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8

    automatic check

  • Attentionlicense conflict noncommercial vs mit

    The LICENSE file is "Attribution-NonCommercial-ShareAlike 4.0 International", where "NonCommercial means not primarily intended for or directed towards commercial advantage or monetary compensation", while setup.py sets `license="MIT"` and the classifier "License :: OSI Approved :: MIT License". GitHub reports NOASSERTION, so no tool can tell which one applies to what you make with it.

    AI review

  • Attentiondevelopment stopped

    The last commit in the checked snapshot is 2024-09-29, 767 days before the check, there are no releases, 75 issues are open, and the last TODO item ("Add the style-transfer and inpainting code for the audioldm_48k checkpoint") is still unchecked with "PR welcomed".

    AI review

  • Attentionpinned old python stack

    The install instructions ask for `python=3.8`, and setup.py pins `transformers==4.30.2`, `numpy<=1.23.5`, `librosa==0.9.2` and `gradio<4`. requirements.txt also sets `--extra-index-url https://download.pytorch.org/whl/cu113`, a CUDA 11.3 build. Those versions are years old and the pinned transformers is the one the checks flagged for known vulnerabilities.

    AI review

  • Infoinstalls unpinned code from git

    requirements.txt starts with `git+https://github.com/huggingface/diffusers.git` and the Diffusers section tells you to run `pip install --upgrade git+https://github.com/huggingface/diffusers.git transformers accelerate`. That installs whatever is on those branches on the day you install, not a fixed version.

    AI review

  • Infooutput quality varies with hardware

    README section "Random Seed Matters": "Sometimes model may not perform well (sounds weird or low quality) when changing into a different hardware. In this case, please adjust the random seed and find the optimal one for your hardware." The CLI also offers `-n` candidates per text as "Automatic quality control".

    AI review

  • Infoheavy on a mac

    README on devices: "mps ( Notice that the computation requires about 20GB of RAM. )" — so on an Apple Silicon Mac you need a machine with a lot of memory.

    AI review

  • Infofaster path lives elsewhere

    README: "The Diffusers version of the code runs upwards of **3x faster** than the native AudioLDM 2 implementation, and supports generating audios of arbitrary length", pointing at the Hugging Face `cvssp/audioldm2` checkpoints. The code in this repository is the slower original.

    AI review

Evidence is shown in its original language.

What Radar could not check about AudioLDM 2

  • Whether the install still works today with Python 3.8 and those pinned versions
  • Whether the two Hugging Face demo pages are still online
  • How the generated sound effects, music and speech actually sound
  • Real disk space taken by the seven model files you may download
  • How long one clip takes on a plain processor, without a graphics card
  • Which licence applies to audio you generate, given the conflict between LICENSE and setup.py

Similar projects

The RepoPicks app

Coming soon: install in 1 click with the RepoPicks app.

Before you download

See the release page on GitHub