Skip to content

CLaMP 3

Cross-modal and cross-lingual music retrieval

Visit official site

What CLaMP 3 is

CLaMP 3 is described as a music information retrieval model that aligns scores, performance signals, audio and multilingual text through contrastive learning. It supports cross-modal and cross-lingual retrieval with generalisation to unseen languages.

What you can do with it

  • Retrieve audio from a musical score
  • Search music with text in several languages
  • Match music from an image description
  • Classify music without labelled examples

Datasets and alignment

  • Features are aligned by contrastive learning
  • Training uses the large M4-RAG dataset
  • WikiMT-X is provided as a benchmark
  • Semantic similarity is used for evaluation

Getting started

  • Open the online demo or download the weights
  • Enter a text, image or other modality query
  • The model aligns the query with music features
  • The closest matching music is returned

Pros & cons

✓ What we like

  • The framework covers several modality pairs
  • The M4-RAG and WikiMT-X datasets are provided

! What to watch out for

  • Retrieval is driven by contrastive feature alignment
  • The stated evaluation is semantic similarity

FAQ

Which retrieval tasks are supported?

The material lists score-to-audio, multilingual text-to-music and image-to-music retrieval.

Which datasets are named?

The feature list names M4-RAG and the WikiMT-X benchmark.

Last reviewed: 2026-11-15

More AI music creation tools

View all →

How we review