Music, Memory and Machine Intelligence

I am glad to announce the start of a new research programme 'Music, Memory and Machine Intelligence' at Tampere University of Technology. I will be running the programme in collaboration with prof. Esa Räsänen from Quantum dynamics research group and prof. Tuomas Virtanen from the Signal processing lab.

One of the main aims of the programme is to test whether novel artificial neural networks are able to produce new musical compositions with long-ranged temporal structures ubiquitous in human composition. To answer this question, we must first invent ways to quantify the rhythmic, melodic and harmonic structures present in various genres and styles of western music. This data will then be used in validation of machine-made musical compositions, to complement usual listening. Network architectures and algorithms will be compared and developed further according to our findings.

Through the research, we will learn to quantify essential statistical features that characterise human composition and, in particular, their evolution in time as the music plays on. These include changes in pulse and meter, accentuation and syncopation, melodic motion, tonality and harmony, all of which contribute in build-up and release of tension in music as experienced by the listener. To compose such music, and to appreciate it, various levels of memory are clearly needed. The same applies to both human- and machine-made composition. Therefore, we will learn not only about the importance and role of various types of memory in human music making but also on their implementation in machine intelligence. The programme will benefit the amateur and professional musician as the tools developed within the programme are made publicly available.

The project was made possible by The Finnish Cultural Foundation's Pirkanmaa Regional Fund.

#music #artificialintelligence #neuralnetworks #machinelearning #memory