flothesof / songcreator

Machine learning based song creator

SongCreator

This is a machine learning based song creator. It follows the principles of the Microsoft Research edited Songsmith (ex Mysong) software, as exposed in [1] and [2]. In this first stage, it can do two things:

• produce some graphical output based on a MusicXML file

  • frequency display for all chord transitions present in the file
  • frequency display of the twelve chromatic notes over the chords that appear in the song

• produce some graphical output based on a folder containing MusicXML files

Usage

Click the Open .mxml file to load a MusicXML file. The two tabs are then updated according to the data from the file.

• The Chord analysis tab displays a bar plot with the chord transitions that appear in the tune and their frequency
• the Melody analysis tab displays a matrix with each row being the frequency of the twelve chromatic notes played over each chord in the song

Note: when clicking the transpose to C button, you'll see the equivalent chords and melody matrices if the song was played in C major.

Open the Folder analysis tab and click on the button Open folder containing .mxml files to parse all MusicXML files in a directory. Two matrices will be displayed in the graphical view.

• On the left, you'll see the chord transition matrix, indexed by chord root. The five rows underneath each root correspond to major, minor, augmented, diminished and suspended chord types.
• On the right, you'll see the melody observation matrix, indexed by chord root in rows and note in columns. This gives the probability of playing a given note over a given chord in the melody.

Bibliography

1. D. Morris, I. Simon, and S. Basu, "Exposing Parameters of a Trained Dynamic Model for Interactive Music Creation.," in AAAI, 2008, pp. 784–791.
2. I. Simon, D. Morris, and S. Basu, "MySong: automatic accompaniment generation for vocal melodies," in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2008, pp. 725–734.

Links

• Snippet for integrating Matplotlib widget into a PyQt4 application
• Snippet for sorting a dict by values

songcreator's Issues

New thesis/project relevant to this one

I've been researching this topic more and last night discovered a recent thesis that takes Songsmith as an inspiration. The author has just posted a first working version of the code here.

cheers,
Wade

Generating Songsmith-compatible model files

I'm trying to use SongCreator to analyze the Wikifonia scores archive for use with Songsmith now, but I can't figure out how to get a result that looks anything like the stock ones (available here ). The files there give log probabilities. Maybe I need to do some sort of conversion?

Assorted errors

Hi, I was hoping to use this program to make new Songsmith chord and melody matrices from this MusicXML collection but I keep getting various errors that I don't know how to address; I figured I'd list them here. The first is that running the folder analysis on this repo's wikifonia folder only shows empty graphs. The second: running melody or chord analysis on one of the included files gives the TypeError message, "setText(self, str): argument 1 has unexpected type 'tuple'. Also, starting SongCreator.py is giving Qt-related messages: "Attribute Qt::AA_EnableHighDpiScaling must be set before QCoreApplication is created."

I'm using Python 3.5.4, by the way, on Windows. When I used Python 3.6.9 I kept getting a different error about how the function that reads a zip file was expecting bytes and not strings.

Difficulties with chord transitions matrix

From the screenshots in the readme it looks like SongCreator treats some chords as repeating--which makes sense, except when I asked the creator of Songsmith why his program does not allow for these "self-transitions," he answered in this thread . Maybe they should be ruled out?

Printing matrices to file?

I'm hoping to create model files that I could try out in Songsmith. Is there any easy way to print the matrices to a text file? It looks like SongCreator creates numpy arrays, but I've never used numpy and I don't know if there's an easy way to print them.