Full text
| |
| Author(s): |
Adriano Claro Monteiro
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Artes |
| Defense date: | 2016-02-29 |
| Examining board members: |
Jônatas Manzolli;
Stéphan Olivier Schaub;
José Henrique Padovani Velloso;
Luis Felipe de Oliveira;
Daniel Luis Barreiro
|
| Advisor: | Adolfo Maia Junior; Jônatas Manzolli |
| Abstract | |
This thesis presents an analysis of the use of spectro-temporal representation of sounds, grounded on auditory neurophysiologic researches, to automatic segmentation of musical recordings. It focuses on the analysis of music in which the musical note concept (i.e the musical note as the minimum element of music construction) is replaced by sounds with a variety of spectral and temporal configurations. The sound representation yield by these models were used as input for a music segmentation algorithm, a standard method from the Music Information Retrieval (MIR) area, based on the identification of temporal intervals with high levels of self-similarity. To evaluate our results we compared them with manual segmentations of music scores, as well as analyzed statistics of this comparison against the ones obtained by segmentation process using other spectro-temporal or instantaneous spectral models. The musical works selected to our analytical experiments were: Innermost Man, by the Russian composer Dmitri Kourliandski, and Pression, by the German composer Helmut Lachenmann. Our results show that the spectro-temporal sound representation based on neurophysiological knowledge had a better performance for the task of automatic segmentation of musical form than the other sound representation models evaluated. This thesis is divided as follows: Chapter 1 presents our objectives, theoretical assumptions, and adopted terminology. Chapter 2 contains a bibliographical review of areas that intersect this work: analytical works of music containing sounds with diversity of spectro-temporal configurations; methods for automatic segmentation of musical form; a brief revision on the physiology of the human auditory system focusing on important aspects related to the computer models adopted; a brief contextualization on the works used in our experiments as well its related composers. Chapter 3 describes our computational methods for audio analysis, and our experimental design. Chapter 4 shows the manual and automatic segmentation of the selected works, as well the statistical evaluation of the results yield by the segmentation algorithm regarding the sound representation models applied. In Chapter 5 we discuss our results and projections to future works (AU) | |
| FAPESP's process: | 13/20528-0 - Application of a computational model of human auditory system for analysis and representation of contemporary music recordings |
| Grantee: | Adriano Claro Monteiro |
| Support Opportunities: | Scholarships in Brazil - Doctorate |