Pop2Piano : Pop Audio-Based Piano Cover Generation

Jongho Choi, Kyogu Lee

Before listening : We highly recommend that you listen to the samples in stereo audio.
All stereo audio has the piano cover on one side and the original song on the other.

Paper Code Colab HuggingFace

Generation Sample
> Dataset Sample
(Fig 4.) on the Paper
Demo Video

Qualitative comparison

This page is for showing examples of PSP dataset and its raw/preprocessed versions.
All song samples on this page are randomly sampled without any curations.

{Pop, Piano Cover} paired dataset should be preprocessed including synchronization, and note quantization.
Because the web-crawled data is sometimes misaligned and human-made piano covers do not follow the original song's progression perfectly.

You can listen to 10 raw samples of PSP below, and you may easily catch what these mean.
Song Is preprocessed

L/R Relative Volume :

(It may take time to load the audio)