Amphion vs AudioCraft
Which should you use? A side-by-side look at what each one does, who it suits, what it costs, how hard it is to set up, and where each falls short.
Open source · MIT
Open-source toolkit for research and building speech, singing, music, and audio generation models.
Open source · MIT
A PyTorch library for researching and running AI audio and music generation models.
The short answer
Choose Amphion if you need…
- Researchers who want reproducible implementations of speech and audio generation models
- Junior engineers learning how text-to-speech or voice conversion systems are built
- Teams experimenting with zero-shot voice cloning or voice conversion
- People who need vocoders or evaluation metrics for audio generation work
Choose AudioCraft if you need…
- Researchers who want working code for text-to-music or text-to-sound models
- Developers who are comfortable with Python and PyTorch and want to build audio generation into a project
- Students learning how neural audio codecs and generative audio models work
- Anyone experimenting with generating short music or sound clips from text descriptions
Side-by-side
| Amphion | AudioCraft | |
|---|---|---|
| Type | Library | Library |
| Price | Free and open source | Free and open source |
| Licence | MIT | MIT |
| GitHub stars | 10.3k | 23.8k |
| Setup difficulty | Advanced | Intermediate |
| How you get it | Clone and build from source | Install with a package manager |
| Main language | Python | Jupyter Notebook |
| Last code update | 8 months ago | 8 months ago |
What each one is
Amphion
Amphion is a Python toolkit for generating audio, including speech, singing voice, and sound from text. It bundles several models (such as MaskGCT and Vevo), vocoders, datasets, and evaluation metrics in one codebase. The project aims to support reproducible research and help newcomers learn how these systems work, and it includes visualizations of classic model architectures.
AudioCraft
AudioCraft is a PyTorch library from Meta's research team for generating audio with deep learning. It bundles inference and training code for models such as MusicGen (text-to-music), AudioGen (text-to-sound), and the EnCodec audio codec. People use it to experiment with generating music or sound from text prompts, or to train their own audio models.
Best for
Amphion
- Researchers who want reproducible implementations of speech and audio generation models
- Junior engineers learning how text-to-speech or voice conversion systems are built
- Teams experimenting with zero-shot voice cloning or voice conversion
- People who need vocoders or evaluation metrics for audio generation work
AudioCraft
- Researchers who want working code for text-to-music or text-to-sound models
- Developers who are comfortable with Python and PyTorch and want to build audio generation into a project
- Students learning how neural audio codecs and generative audio models work
- Anyone experimenting with generating short music or sound clips from text descriptions
Strengths
Amphion
- Covers many tasks (TTS, SVS, VC, SVC, TTA, vocoders, evaluation) in one toolkit
- Released under the MIT license and actively updated, with the last push in March 2026
- Aims explicitly at education and reproducibility, with architecture visualizations
AudioCraft
- Bundles several models (MusicGen, AudioGen, EnCodec, MAGNeT, AudioSeal and others) under one library
- Code is released under the MIT license and the project is actively maintained by Meta's research team
- Includes training code and documentation for several of its models, not only inference
Not ideal for
Amphion
- People who just want a finished voice app with no setup, since the README describes a research toolkit rather than an end-user product
- Users without a Python and deep-learning environment, because the project requires running code and models locally
- Projects needing production-grade guarantees, since the latest release is labeled v0.1.1-alpha
AudioCraft
- People who want a finished point-and-click music app: the README describes a library, not a consumer product
- Commercial projects needing to use the pretrained model weights: they are released under CC-BY-NC 4.0, which restricts non-commercial use, while only the code is MIT-licensed
- Users without Python or GPU experience, since installation requires PyTorch and specific versions
Trade-offs
Amphion
- Latest release is an alpha version (v0.1.1-alpha) with no binaries, so expect rough edges
- Several models and datasets have different licenses; for example, part of Emilia is CC BY-NC 4.0
- Text-to-music is listed as still in development in the README
AudioCraft
- Pretrained model weights use a non-commercial license (CC-BY-NC 4.0), which limits commercial use
- Requires specific versions (Python 3.9, PyTorch 2.1.0 per the README), which can complicate setup
Getting started
Amphion
Clone the repository
git clone https://github.com/open-mmlab/Amphion.gitAudioCraft
Install PyTorch 2.1.0
python -m pip install 'torch==2.1.0'Install setuptools and wheel
python -m pip install setuptools wheelFrequently asked questions
What is the difference between Amphion and AudioCraft?
Amphion: Open-source toolkit for research and building speech, singing, music, and audio generation models. AudioCraft: A PyTorch library for researching and running AI audio and music generation models. Amphion suits researchers who want reproducible implementations of speech and audio generation models, while AudioCraft suits researchers who want working code for text-to-music or text-to-sound models.
Are Amphion and AudioCraft free?
Amphion: Free and open source. AudioCraft: Free and open source.
Which is easier to set up, Amphion or AudioCraft?
AudioCraft is easier to get started with (rated intermediate); Amphion is rated advanced and usually needs more setup.
Is Amphion or AudioCraft more popular?
AudioCraft has more GitHub stars (23.8k) than Amphion (10.3k). Popularity isn't the same as fit — check which one matches your use case above.
What are other alternatives to Amphion and AudioCraft?
Other options people compare them with include Bark.
Keep comparing
Built from our catalogue entries for both products: open-source details come from each project's README (AI-assisted, refreshed weekly) and GitHub data; AI-tool details are editorial. Spotted something wrong? Let us know from either product's page.