my picture

Dan Gang


Welcome!!

I was born in Dec. 10, 1962 in Haifa, in Israel.

I am interested to know how people acquire their tacit knowledge on music, and how do they learn to perform in real time so well, complex musical activities such as listening to music.
For this aim, I take ideas from three disciplines: Artificial Intelligence(AI), Music and Cognition. I develop a modular approach in building recurrent neural network (RNN) architectures for real time performance of music. The publications below describe how we apply these architectures for on-line learning of: chord sequences or harmonizing melodies in real time performance situation; modeling listener's expectations: ambiguity, surprises, and meter recognition. We also built a hybrid system that integrates recurrent net with DAI agent-based modules, applied to composing polyphonic music in real time and for representing musical knowledge.

Currently, I am Fulbright Post-Doctoral Fellow at Stanford University at the Center for Computer Research in Music and Acoustics CCRMA . My Ph.D research was completed at the Hebrew University in Jerusalem at the Institute of Computer Scienc in the AI lab.

If this subject seems to be interesting, I would like to refer you to the course: Music, Cognition and Artificial Intelligence (Hebrew University, 1997); and to the course: Modeling Creativity (Stanford University-CCRMA, 1999) which were tought by me.


To read my publications, please follow the links by clicking on the name of the papers below:


More on Melody Harmonization in Real Time with Recurrent Neural Networks

Tuning a Neural Network for Harmonizing Melodies in Real-Time , with: Daniel Lehmann and Naptali Wagner. in Proceedings of the International Computer Music Conference, Michigan, Ann Arbor, October 1998.

Following is a typical representative example which results from our neural network model. The model produces hundreds of results and the following result is not nessecarily the best, from the aesthetic point of view. We use this example to demonstrait some common features that are often found in the many of the resulted harmonization.

The net learns relations between important notes of the melody and their harmonies and is able to produce harmonies for new melodies in real-time. The net ability to generalize is examined by providing it unfamiliar melodies' notes to harmonize in real-time (Swanee River is an example from the generalization set). Real-time harmonization here means, harmonizing without advanced knowledge of the continuation of the melody. I would like to refer you to the section Musical Results, of the paper above, for more detailed and analysis of the results.

Musical Results

Score

In the score of the song Swanee River the upper harmonization is the output of the neural network, the lower found in the book.

Audio Example of the source

To hear an audio example of the harmonized melody of the source book follow the link- Swanee River in wav format (source) .

The neural network harmonization

The net real-time harmonization results in comparison with the source harmonization found in the book:

  • The 4 initial measures that were produced by the net for Swanee River (wav format measures 1-4)are in double harmonic rhythm.
  • The B note in measure 9 interpreted as non-chord note (non-concurrent harmonization), such interpretations are quite rare Swanee River (wav format measures 9-10).
  • The G note of the melody of measure 15 interpreted as chord note of G7 (while in measure 7 it is interoreted as chord note of C chord!). This reveals the ability of the net to be sensitiev for more global hierarchical structure Swanee River (wav format measures 15-16).
  • The neural network harmonization in real-time ofor the whole song Swanee River (wav format measures 1-16).


  • I would like to invite you to explore the world of MP3 format with the music of my band named the GANG band. I put here some of the songs of this new Israeli Rock band. Be aware that all the songs are copyright protected. If you like/hate the music or if you have any remarks please feel free to send them to me.

    GANG band in mp3 format

    List of Songs (partial)

    Mechabek Ota (Give a huge to her)

    Notsa Baruach (Feather in the wind)

    At Yacholt laset(You could stand it)

    © 1998, Dan Gang ;dang@ccrma.stanford.edu>, All rights reserved.
    Some interesting URLs of Computer Music Activities: ICMA, CCRMA, NICI, Litefoot project.
    Some interesting URLs of incoming conferences: ICMC99, AAAI 1999 Spring Symposium Series, NNSP'99, Symposium on AI and musical creativity.
    If you have any comments, you can send mail to me.


    Dan Gang

    Phones:1-650-723-4971 ex. 350(w) 1-650-322-2151(h)
    Fax: 1-650-723-8468
    Email: dang@ccrma.stanford.edu
    http://www-ccrma.stanford.edu/~dang

    Address at Stanford:

    CCRMA, Department of Music
    Stanford University
    Stanford, CA 94305




    Last modified in August, 30, 1999.