Caracterização de ataques na clarineta via modelagem e decomposição espectral

Abstract

Note attacks have an important role in musical performances. By manipulating note attack quality, musicians are able to control timbre, articulation, rhythm, that are essential parameters for conveying their expressive intentions making music more appreciable. Models for musical expressiveness are proposed and improved since the first objective investigations on performances at the beginning of the last century. Modeling aims at a better understanding of how these parameters related to expressiveness are manipulated. In the literature, while the attack's importance for the perception and the performance construction is pointed, the analysis difficulties are also exposed. They are short-duration acoustic events, subjected to interactions with environmental reverberation, with transient and nonlinear characteristics. Studies about note attacks often create conditions that are far from performance, analyzing isolated notes and recording in anechoic chambers to isolate variables. They bring essential results for the construction of knowledge. However, this distance disregards the interaction between the musician and environment, while adapting the performance to the specific acoustics of a concert hall is part of the routine of professional musicians. Relatively recent studies investigating performance adjustments under different room acoustic conditions have revealed results that corroborate that professional musicians consciously adjust their style under different concert halls. In addition, most of the studies found in the bibliographic review carried out that analyze attacks and transitions between notes, use the total energy of the signal, proposing descriptors such as Legato Index, Inclination, and Attack Time Log. Few studies have been found based on spectral content analysis.

Motivated by these limitations and aiming to overcome them, the research presented in this thesis proposes and explores the use of modeling and spectral decomposition of the acoustic signal, as a way to describe and analyze note attacks in clarinet recordings under daily practice conditions. By obtaining the modeling and decomposition of the harmonics in the frequency spectrum, three representations are proposed for the attack region in this instrument: (i) the proportion of the harmonic energy, isolated from the residue, during the transition between notes in which the attack is contained (ii) considering independently the harmonic content of the note being played, the reverberation of the previous note and the residues that are part of the attack but not from the harmonic structures; (iii) considering the temporal profile of the modeled harmonics onsets. In the first representation, a legato index already proposed in the literature is adapted to use the energy of the harmonic component of the transition, instead of the total energy, named Harmonic Legato Index (ILh). In the second representation, the energy proportions of the three components (in relation to the original signal) extracted during the attack were confronted each other in only two dimensions (2D). In this 2D representation (called H2R2 plane), it was possible to visualize and make inferences about the proportions of the harmonic, non-harmonic components and interactions with the acoustics of the room, aiming to enrich the study of musical performances in everyday practice condition, where reverberating environments are always present. In the third front, it is proposed a three-dimensional descriptor, named Harmonic Attack Profile (PHA), involving time, frequency, and spectral magnitudes, in order to characterize the harmonic structure of note attacks on the clarinet. PHA was defined as the sequence of the harmonic onsets, starting with the fundamental frequency, in the increasing harmonic order. Each harmonic onset was considered as 50% of its magnitudes range in dB during the attack.

The ILh, the H2R2 plane, and the PHA were tested using recordings of a clarinet piece from the traditional classical repertoire, played by six professional musicians. For the analysis of the H2, R2, and ILh energy proportions, two groups of notes were selected. statistical tests ANOVA (Analysis of Variance) indicated significant differences (p < 0.05), considering “musician” and “note” as factors. The “condition” factor (with and without a metronome) did not present significant differences (p > 0.05) in the t-tests for any variable. For the H2R2 plane and for the PHA, note attacks were examined in transitions with different technical execution demands, in legato and articulated conditions. In the H2R2 plane analysis, statistical tests for comparison between two-dimensional means (MANOVA - Multivariate Analysis of Variance) indicated significant differences (p < 0.05), considering “musician” as a factor. For the PHA, we evaluated its capacity for classification by musicians, and as it is a three-dimensional profile, the verification was made in the Harmonic-Magnitude and Harmonic-Time projections, separately. The classification capacity was evaluated through comparisons using Pearson's Correlation Coefficient between pairs of profiles in two situations: Intra and Inter classes. The Intra class groups and the Intra and Inter class means were compared using ANOVA and t-test. The tests showed Intra class differences (p < 0.05) in all profiles of harmonic onsets (PHA) of the seven-note attacks evaluated. Significant differences (p < 0.05) between Intra and Inter correlation means occurred in four notes, with three of them to the HxT projection in legato notes, suggesting an indication that the time information of harmonic onsets is more relevant in legato notes. High values of the Inter means for HxM projections (the same does not happen with HxT) suggest the existence of a harmonic magnitude structure composing the PHA, common to musicians, which may be related to the harmonic energy proportions that define the timbre of the instrument.