Project Goals

Audiveris is an Optical Music Recognition (OMR) module. Starting from the image of a music sheet, it provides high-level logical music information compliant with the MusicXML definition. Other tools such as a Midi Sequencer, or a Composition Editor can then read and update this standard data.

There are already commercial tools in this area but Audiveris is, to our knowledge, the first Java open-source OMR tool. It is a cross-platform tool, written entirely in Java, and tested on Windows, Solaris, Linux and Mac OS (Mac test is on-going).

Audiveris works with printed music sheets only, the task of recognizing hand-written scores being significantly harder.

Current Status

Starting with version 3.0, Audiveris is now a fully functional music scanner. Since there may still be remaining errors in the translation, the program offers the user the ability to guide the translation on the fly. It can also take advantage of user-supplied data in order to further train its internal neural network, and thus improve the glyph recognition task.

More detailed information is available in the provided manuals, but here are the main phases of the application:

1. Detection and correction of potential skew angle of the sheet,
2. Removal of staff lines, detection of systems, staves and measures,
3. Extraction of individual music items, essentially based on their appearance and their structure,
4. Music translation based on the extracted music items, and essentially driven by music grammar,
5. Manual correction of unavoidable remaining errors,
6. Export of logical music information, using the MusicXML format definition,
7. Finally, any other application able to read / write MusicXML format can be chained to it.

There is still room for improvement in this program but the internal architecture is now fairly stable and robust enough to be used and augmented by people willing to contribute.

Here are a few ideas of contribution that would make sense and be really appreciated:

• Rather easy
• Validate Audiveris on other platforms than Windows, Solaris or Linux. Some people volunteered for Mac OS (Brenton Partridge is actively working on it).
• Integrate with a MusicXML player such as Xenoage of Andreas Wenger.
• More significant
• Find and integrate an open-source OCR (Optical Character Recognition), preferably in Java, in order to handle the various text pieces (typically lyrics) which for the time being are simply discarded.
• Integrate with a Visual Music Editor for manual correction, for printing, etc. Audiveris has only an embedded rudimentary editor. For example Xemo would be a natural partnering tool, however the Xemo site doesn't seem active anymore.
• Evaluate a move to NetBeans (the platform, not the IDE). This would move Audiveris stand-alone application to a much better integration environment, taking advantage of view and parameter management. This could also ease the integration with visual editors.
• Fairly difficult
• Consider addressing hand-written scores. This domain could take benefit of a significant part of Audiveris framework, however this would really be a project on its own. To be discussed if anyone is interested by this daunting task.

Download

Documentation

All documentations are bundled with the source archive above. They are also available on-line here:

Implementation

Audiveris is a pure Java stand-alone application (about 260 classes for 85000 lines of Java code)

• Techniques
  • Linear Adjacency Graphs (LAG) record pixel information as a directed graph structure, composed of pixel contiguous runs organized in sections.
  • Sticks are developed above LAGs to take advantage of the structuring sticks of a music sheet (staff lines, ledgers and legato signs are such horizontal sticks, while stems and barlines are vertical sticks).
  • A Neural Network allows efficient pattern recognition as well as training based on user input for continuous improvement of the tool efficiency.
• Technologies
  • Java Advanced Imaging (JAI) for number crunching in image processing (loading image, accessing pixel data, rotating the image, etc)
  • Swing or user interface, with JGoodies enhancements in terms of forms and look & feel.
  • JFreeChart for charts.
  • JAXB 2 for XML persistency.
  • Jama for linear algebra in Java.
  • Java 6 is the preferred choice for Java, since JAXB 2 is already integrated in this distribution.
• Companions
  • ProxyMusic is a separate project, dedicated to the marshalling between Java objects and MusicXML files. Audiveris today supports MusicXML V1.1 and will soon support MusicXML V2.0 which Recordare has just released. Please refer to https://proxymusic.dev.java.net