Automated Phrase Analysis of Sonatas

Abstract

A musical phrase is "the most basic unit” of a musical form In the field of music information retrieval (MIR) there has been numerous research about how to segment the phrases of monophonic and polyphonic However the phrases of the homophonic is rarely discussed This study is based on the sonatas in the 18th and 19th centuries and proposes an automatic tool to analyze harmony tonality cadence rhythm and phrase The sonata is the most common and artistic representation of the musical form in the period of Classicism As a tonal music the melody and the harmony have clear directionality and the cadence is their regular objective Composers often use modulation to let different tonal segments be sharply distinct Various kinds of cadence are used in the modality Some can properly give the music a perfect end but others can be quite unstable that the music seems to go forward A cadence always has a longer note to let the music slower; a phrase usually ends with a cadence This study makes progress on the chord recognition algorithm of the template matching proposed by previous studies Moreover we can distinguish the tonality by using diatonic scale and diatonic chords Finally the position where the cadence should be is inferred from the combined features of the tonality the chord and the note In addition to the observation on the tonality and the harmony phrases can also be analyzed by repetition and variation in rhythm This study separately analyzes the rhythm complexity of the sonata’s left and right hands The right hand is defined as the melody the outer voice of the musical composition and its rhythm changes frequently and complicated The left hand is defined as the accompaniment providing the melody a stable harmony to support or response the rhythm is unchangeable and simple After quantifying the complexity of the left and right hand rhythm they are divided into groups via continuity and we can combine this with the cadence to find out the boundary of the phrase According to the data we selected four composers Haydn Mozart Clementi and Beethoven from the Classicism Then we choose two sonatas with the first movement (sonata form) of each composer The entire chord the tonality the cadence the phrase and the period are manually annotated through two groups of musicians Comparing with the previous study we put more effort on phrase analysis than cadence analysis In order to evaluate more accurately we use the unit of "beat" instead of "bar" The results show that our algorithm can provide preliminary segmentation of phrases In the total of 1283 bars the F-Measure of the 512 phrases is 41%

Date of Award	2019
Original language	English
Supervisor	Wen-Yu Su (Supervisor)