Automatic characterization and generation of expressive ornaments from bassoon audio recordings

Expressive performance characterization analyzes differences in performances, performers, playing styles and emotional intentions. Most research focus on studying timing deviations (Dixon2005, Honing2002), dynamics and vibrato (Desain1999). They often analyze MIDI performances and sometimes recordings (Dixon2005, Gomez2003). This characterization is used later to generate expressive performances from score (Sundberg2003).

Less research is devoted to ornamentation. Ornaments are indicated in the score, without any explicit information about timing and dynamics. Some works have studied piano performances (Moore1992, Palmer1996, Brown2003, Timmers2002).

The main goals of this work are first, to study the behaviour of ornamentation by analyzing timing and dynamics from bassoon recordings. Then, the acquired knowledge is used in order to generate expressive trills in MIDI format.

We divide our study in two main areas
1. Characterization of a set of expressive recordings (Sonata from Michele Corette) played by a professional bassoon performer. Each movement is played in three different tempi, obtaining 96 ornaments (trills and appoggiaturas). A melodic description is obtained for each ornament. We then perform some statistical analysis to model their behaviour, and then use some machine learning techniques.
2. Generation of expressive ornaments, testing different machine learning methods. Given a melody with the indicated ornaments, the system generates all the notes within the ornaments.

The melody estimation is successfully adapted to the particular analysis of bassoon ornaments. The statistical analysis reveals a similar behaviour to previous studies on piano. The speed of execution is around 8 notes per second for most of the trills. We can distinguish that the first and the last notes are usually longer than the central ones for slow tempi. For fast tempi, trills are usually converted into appoggiaturas. We finally identify regularities in the execution of central notes. We finally present examples of automatically generated, giving promising results.

This study presents an approach for the automatic analysis and generation of expressive ornaments of bassoon. Further work is centred in increasing the analyzed collection in order to obtain a robust model and to extent it to other musical instruments.