Tools for the analysis of alterations in vocal musical performance

Dimitrij Hmeljak, Domenico Zanon
Dipartimento di Elettrotecnica, Elettronica ed Informatica
Universita' di Trieste, via A.Valerio 10,  34127 Trieste, Italy
Fax: +39-40-6763460

Abstract


In this paper we approach the study of un-voluntary pitch alterations 
in the singing voice.
It is known that the richness of the human voice expressivity is 
partly achieved by alterations of the pitch outside of the tempered-
scale. An accurate tool for the analysis of such irregularities is 
necessary for the understanding of their influence on the global 
performance of a singer.
We developed a software system written in C language, based on the 
Apple Macintosh platform, to analyse some algorithms for pitch-
detection. Two well-known algorithms used in non-musical applications 
were adapted for the analysis of a singing voice: 1) a combination of 
Fast Fourier Transform and Average Magnitude Difference Function 
methods and 2) the Fundamental Period Measurement by an array of IIR 
Filters. Both output data were processed to obtain two different 
analysis possibilities: 1) on-screen drawing of the pitch line on a 
pentagram and 2) extraction of MIDI-data for post-processing by the 
Finale notational software. The developed system was compared with an 
experimental tool for automatic transcription of melodies available in 
the Logic Audio software package by EMagic, and some results will be 
described in the final paper.
The comparison showed that our system, a tool focused exclusively at 
the study of this problem, if accurately tuned, can be as effective as 
more expensive complete commercial solutions.