Lisztomaniaaa/piano_trans

Intro

Qiuqiang Kong High-Resolution Piano Transcription System of ByteDance is an advanced music information retrieval tool specifically designed to automatically convert audio signals from piano performances into detailed sheet music with high accuracy. The system's training model mirrors the latest deep learning techniques, including convolutional neural networks and recurrent neural networks, to achieve accurate capture of note timing and pitch. Through multi-scale feature learning and modeling of long-term dependencies, the system is able to handle complex musical structures and provide accurate transcription of high-density notes. This ByteDance system not only optimizes the efficiency of music analysis and theoretical research, but also provides powerful technical support for music education and performance.

Demo

<https://huggingface.co/spaces/Genius-Society/piano_trans>

Usage

from huggingface_hub import snapshot_download
model_dir = snapshot_download("Genius-Society/piano_trans")

Maintenance

GIT_LFS_SKIP_SMUDGE=1 git clone git@hf.co:Genius-Society/piano_trans
cd piano_trans

Mirror

<https://www.modelscope.cn/models/Genius-Society/piano_trans>

Thanks

• —High-resolution Piano Transcription with Pedals by Regressing Onset and Offset Times

Provenance

This repository is a duplicate of `Genius-Society/piano_trans`, forked to serve as the base checkpoint for further fine-tuning/retraining.

• —Original model: Genius-Society/piano_trans
• —Original architecture/method: Kong et al., "High-resolution Piano Transcription with Pedals by Regressing Onset and Offset Times" (arXiv:2010.01815), ByteDance.
• —License: MIT (preserved from the original repo; see LICENSE).