Skip to content

Make-An-Audio 2

Text to audio generation with a diffusion model

Visit official site

What Make-An-Audio 2 is

The technique is described as a diffusion model that turns text into audio with help from pre-trained large language models.

What you can do with it

  • Parse input text with a pre-trained LLM
  • Learn semantic alignment with a structured encoder
  • Generate audio through the diffusion denoiser

技術の要点

  • A feed forward Transformer denoiser handles variable length audio
  • LLMs convert audio label data into an audio text set
  • Time information is drawn from the parsed text

Pros & cons

✓ What we like

  • Variable length audio generation is addressed by the denoiser
  • LLMs are used to ease the shortage of temporal data

! What to watch out for

  • The described input is natural language text
  • The generation flow runs through several encoder stages

FAQ

Who developed it?

The material names researchers from Zhejiang University, ByteDance and the Chinese University of Hong Kong.

What does the structured encoder do?

The feature list says it helps learn semantic alignment in the denoising process.

Last reviewed: 2026-11-15

More AI music creation tools

View all →

How we review