ìConsistency of Timbre patterns in Expressive Music Performanceî


Mathieu Barthet, Richard Kronland-Martinet, Solvi Ystad


International Conference on Digital Audio Effects (DAFx-06)

Montreal, Canada, September 18-20, 2006



Sound examples:



A professional clarinettist was asked to play a short piece of music (the beginning of a Bachís Cello Suite) twenty times with the same musical intention.



Figure 1 : Score


Please click on the following link to play one of the recorded performances: P15


Strong changes of spectral centroid can be observed within the duration of the notes. For example, here is the spectral centroid of the same note N5 (the fifth note of the melody which is a B) for several performances (P15 to P19).




Figure 2 : Spectral Centroid of the fifth note for different performances



For each performances, the spectral centroid of the note globally increases within its duration. This systematic behaviour might be closely related to musical interpretation. One can hear the brightness of the note N5 getting higher within its duration. Please click on the links below to play or download the sound files corresponding to the five performances P15 to P19.

















We applied signal transformations to the original note to point out these changes of brightness. For that purpose, we forced the spectral centroid to remain stable from the end of the attack part until the beginning of the decay part. The transformation keeps the original attack, decay and intensity envelope of the note. We made two sounds, one for which the spectral centroid is that of the original note close to its beginning (low brightness), and one for which the spectral centroid is that of the original note close to its end (high brightness).



Original note


fixed Spectral Centroid (close to the beginning of the note)


fixed Spectral Centroid (close to the end of the note)





Figure 3 : Freezing the Spectral Centroid



Remarks about the sounds:


- The sampling frequency is Fe = 44.1 kHz.

- The sound P15 corresponds to the original recorded performance.

- The other sounds have been produced by using an additive synthesis model with 50 harmonics.