Machine_Learning_for_Chord_Recognition

Team 201612-44 for Big Data Analytics
Reva Abramson ra2659 CVN
https://www.youtube.com/watch?v=BucCPu1yZrA

##Overview Without an ear for music it is difficult to play the songs you like correctly. This Spark and Python project uses machine learning to recognize chords.

##Usage

Getting the Data

• Download annotated (labled with chords) data from isophonics.net (*.lab files)
• Find the matching audio file on youtube and convert it to mp3 format and name it the same (except for the extension)
• put all this data in a music directory
• An example is hello-goodbye.lab and the corresponding hello-goodbye.mp3 in this repository

Preparing the Feature Vectors

• install Spark over Hadoop
• install the Python library librosa pip install librosa
• install ffmpeg brew install ffmpg
• run features.py and direct the output to a file
• the results should be in the format of the file output.txt which is what Spark expects

Creating the Model

• run train.py in the same directory as the output.txt file
• a model will be created called rf.model

Predicting with the Model

• run chordviewer.py in the same directory as the mp3 song you want to predict and in the same directory as the model rf.model
• hardcode the name of the song in chordviewer.py line 41, or alternatively name your song upload.mp3

###Setting up the Web Server and CGI Application

• on a mac run sudo apachectl start in order to start up your apache web server at localhost
• navigate to /private/etc/apache2/httpd.conf and uncomment #loadmodule cgi
• move the model rf.model and chordviewer.py into /Library/WebServer/CGI-Executables
• move index.html into /Library/WebServer/Documents
• navigate to localhost, upload a song, and the chords will be predicted!