music_transcription_MAPS-master
所属分类：人工智能/神经网络/深度学习
开发工具：Python
文件大小：87375KB
下载次数：17
上传日期：2019-05-30 07:17:11
上 传 者：
小章
说明： PyTorch机器学习音乐分类，导入wav和txt进行机器学习，运行项目的命令行和需要的音乐文件都在在压缩包里，项目可以直接运行。
(PyTorch machine learns music classification, imports wav and TXT for machine learning, runs the command line of the project and the required music files in the compressed package, and the project can run directly.)
文件列表:
AkPnBcht (0, 2019-05-28)
AkPnBcht\MUS (0, 2019-05-28)
AkPnBsdf (0, 2019-05-28)
AkPnBsdf\MUS (0, 2019-05-28)
AkPnCGdD (0, 2019-05-28)
AkPnCGdD\MUS (0, 2019-05-28)
AkPnStgb (0, 2019-05-28)
AkPnStgb\MUS (0, 2019-05-28)
appendix (0, 2018-12-05)
appendix\MAPS_MUS-chpn_op66_ENSTDkAm.png (74462, 2018-12-05)
config.py (469, 2018-12-05)
data_generator.py (1129, 2018-12-05)
ENSTDkAm (0, 2019-05-28)
ENSTDkAm\MUS (0, 2019-05-28)
ENSTDkCl (0, 2019-05-28)
ENSTDkCl\MUS (0, 2019-05-28)
features (0, 2019-05-28)
features\logmel (0, 2019-05-28)
features\logmel\06-DieSonne_SptkBGAm.p (1680189, 2019-05-28)
features\logmel\07-HerrGott_SptkBGCl.p (1642929, 2019-05-28)
features\logmel\08-FuerDeinenThron_StbgTGd2.p (1676949, 2019-05-28)
LICENSE.txt (1099, 2018-12-05)
main_dnn (0, 2019-05-28)
main_dnn.py (11789, 2019-05-28)
main_dnn\logmel (0, 2019-05-28)
main_dnn\logmel\md_1000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_2000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_3000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_4000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_5000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_6000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_7000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_8000iters.tar (10672357, 2019-05-28)
main_dnn\logmel\md_9000iters.tar (10672357, 2019-05-28)
packed_features (0, 2019-05-28)
packed_features\logmel (0, 2019-05-28)
packed_features\logmel\test.p (22, 2019-05-28)
packed_features\logmel\train.p (4999993, 2019-05-28)
prepare_data.py (17684, 2019-05-28)
... ...
# Automatic music transcription (AMT) of polyphonic piano using deep neural network (Implemented using pytorch).
## The "Learning the Attack, Decay and Pitch for Piano Music Transcription" code will be pushed soon.
Author: Qiuqiang Kong (q.kong@surrey.ac.uk)
## Summary
A fully connected neural network is used for training followed [1] (implemented using pytorch). Log Mel frequency with 229 bins are used as input feature [2].
## Result
On the test set, frame wise F value around 75% can obtained after a few minutes training.
## Dataset
Download dataset from http://www.tsi.telecom-paristech.fr/aao/en/2010/07/08/maps-database-a-piano-database-for-multipitch-estimation-and-automatic-transcription-of-music/
If you fail to download the dataset, you may download the already calculated log Mel feature & ground truth note from here https://drive.google.com/open?id=1OtK4tSrparkYVg_IrQvSDPJRtlwaQ_1k
## Install requirements
1. pip install -r requirements.txt
2. Install pytorch following http://pytorch.org/
## Run
1. Modify dataset path in runme.sh
2. Run ./runme.sh
## Transcription result of Chopin Op. 66 Fantasie Impromptu
Real play: [real_play.wav](https://drive.google.com/open?id=1kwhsM2b_CmPfnRgJPqPmtCRn9bIm7qoD)
Transcripted result: [midi_result.wav](https://drive.google.com/open?id=1HwnVdPZjRxqNE-hum1FLxyZTfjaK8P29)
Visualization of piano roll.
## Reference
[1] Sigtia, S., Benetos, E. and Dixon, S., 2016. An end-to-end neural network for polyphonic piano music transcription. IEEE/ACM Transactions on Audio, Speech and Language Processing (TASLP), 24(5), pp.927-939.
[2] Hawthorne, C., Elsen, E., Song, J., Roberts, A., Simon, I., Raffel, C., Engel, J., Oore, S. and Eck, D., 2017. Onsets and Frames: Dual-Objective Piano Transcription. arXiv preprint arXiv:1710.11153.
近期下载者：
相关文件：
收藏者：