MPhil Thesis Defence


"Wave-to-MIDI Transcription of Polyphonic String Music"

By

Mr. Yiu-Wai Fung


Abstract

Automatic music transcription is one of the most active areas in recent
computer music research. Music transcription systems are useful in a wide
range of applications including melody extraction, musical database query
and interactive music systems. Wave-to-MIDI transcription, being a subset
of full automatic music transcription, is a challenging problem to solve
by itself. This thesis proposes a Wave-to-MIDI transcription system which
is specialized for polyphonic string music, where the input musical
signals are converted from acoustic to symbolic representation.

The proposed system is divided into two stages. A sinusoidal analysis is
performed in the analysis stage. Spectral peaks are extracted with
parabolic interpolation and matched across frames to form tracks. In the
note detection stage, different smoothing techniques are applied to the
detected tracks. Inharmonicity and string spectrum characteristics are
incorporated to enhance harmonic detection. Harmonic penalization is also
performed to minimize the effect of false octave hits. A MIDI
transcription with all detected notes is the result. Experiments show
encouraging results for the transcription output using window-based and
track-based quantitative analysis. A high percentage of notes are detected
by the system for various monophonic and polyphonic inputs. Listening test
results also indicate a significant improvement over a previous
transcription system.


Date:				Tuesday, 18 July 2006

Time:				9:00a.m.-11:00a.m.

Venue:				Room 4502
				Lifts 25-26

Committee Members:		Prof. Andrew Horner (Supervisor)
				Prof. Vincent Shen (Chairperson)
				Dr. David Rossiter


**** ALL are Welcome ****