Transcriber is a software system that automatically transcribes audio data from solos or duets into MIDI scores. This software is still being developed.

The front-end of the system analyzes audio files using MatLab scripts and produces analysis files. The SMS Toolkit (X. Serra), AuditoryToolbox (M. Slaney), SMT (K. Johnson), and Voicebox were all used. The MatLab scripts that were used for the front-end are available here:


The back-end of the system consists of a Java program that constructs various neural network configurations that can be trained on the output of the front-end. Analyses from the front-end can then be processed through the trained networks, which produce MIDI scores. The back-end of the Transcriber software is written in Java. If your system does not already have the Java Runtime Environment installed, it can be downloaded for free from the web site. Click on the link below to download a zip file containing the Transcriber back-end. A README.txt file is included that gives installation instructions and an overview of how to use the software. The java source files are also included, for those interested in seeing how the image processing techniques were implemented.


-top of page-