Skip to content
Get packs

Amphion

Open-source toolkit for research and building speech, singing, music, and audio generation models.

GitHub Download v0.1.1-alpha
Stars
10.3k
Forks
855
Licence
MIT
Language
Python
Last update
8 months ago
Type
Library

What it is

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.

  • Run text-to-speech or voice conversion experiments using the models listed in the README
  • Compare evaluation metrics across audio generation outputs
  • Study model architectures through the project's visualizations
  • Build or curate large-scale speech datasets using the dataset tooling referenced in the README

Great for

  • 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

Not ideal for

  • 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
Step-by-step guide: how to use AmphionInstall, first run, examples, tips and FAQs
Amphion vs BarkAmphion vs AudioCraft

How to get it

  1. 1

    Clone and build from source

    You'll need: Python · Git

    Clone the repository

    git clone https://github.com/open-mmlab/Amphion.git
  2. 2

    Run it

    • Clone the repository and open the folder.
    • Read the README section for the task you want (for example the TTS or SVC guide linked from the README badges).
    • Follow the environment and model setup instructions in that task's README.
    • Run the example recipe for that task and check the output audio.
  3. 3

    Get the most out of it

    • Start from the task-specific READMEs linked in the top badges (TTS, VC, SVC, TTA, Vocoder, Evaluation).
    • Use the pre-trained models published on Hugging Face or ModelScope, as linked in the README.
    • Try the visualizations of classic architectures to understand models before modifying them.
    • Join the Discord community linked in the README for questions and updates.

Downloads

Source code (.zip)main branch · straight from GitHubRelease v0.1.1-alphaRelease notes and sourceOfficial website & docsopenhlt.github.io/amphion/

Strengths

  • 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

Trade-offs

  • 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

How it compares

ProjectStarsTypeLevelUpdatedBest for
Amphion (this)10.3kLibraryadvanced8 months agoResearchers who want reproducible implementations of speech and audio generation models
bark39.3kAI model / agentintermediate2.2 years agoDevelopers who want to add text-to-speech or sound generation to a Python project
audiocraft23.8kLibraryintermediate8 months agoResearchers who want working code for text-to-music or text-to-sound models
piper11.3kAppintermediate1.2 years agoDevelopers who want offline speech output in a local app or home automation setup
leon17.6kAppintermediate4 days agoDevelopers who want to inspect or extend an open-source assistant
GPT-SoVITS64.6kAppintermediate5 days agoHobbyists who want to experiment with cloning a voice they have permission to use

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.