National Technical University of Athens (NTUA)
Αυτόματη ανάκτηση μουσικής πληροφορίας με έμφαση στο ρυθμό
Abstract
dc:descriptionThe purpose of this thesis is to develop techniques for automatic analysis of musical signals. Great emphasis is placed on automatic analysis of rhythm, which is a fundamental characteristic of music, as it describes the temporal structure and organization of the music signal. Rhythm is a structural element in an automated music transcription system, and the rhythmic information can be used in other important applications such as indexing and retrieval of music content calculation of rhythmic similarity and mixing of music signals.Paramount importance to the development of an automated analysis system of rhythmic content is the periodicity function. The periodicity function is the "rhythmic spectrum" of a music signal, as it demonstrates the salience of different targeted periodicities. A large part of this thesis is dedicated to the extraction and processing of a periodicity function.Using modern signal processing techniques such as source separation, we propose a periodicity function, which is a robust representation of rhythmic content. Then we apply unsupervised learning techniques on the periodicity function for extracting solid rhythm features. These features are used in two problems of rhythmic categorization, the automatic categorization in rhythmic classes and the extraction of time key. We propose three different tempo extraction methods based on the periodicity function and the extracted features, as well as a method for beat tracking.Furthermore, the periodicity function is redefined so that it can be approximately reversible, i.e. it is possible to reconstruct a music signal from the periodicity function that maintains the rhythmic structure of the original signal.Finally, the rhythmic features are extended with the incorporation of timbral and harmonic features in order to build a content-based music similarity system. This system was integrated in a content based music search web platform.In summary, this thesis deals with six distinct music analysis problems, namely, music key extraction, dance style classification, tempo estimation, beat tracking, and content based similarity between music tracks. Furthermore, it proposes a method for the calculation of a "reversible" periodicity function. The proposed methods were evaluated for all the problems in a wide range of data sets and compared with other state of the art methods achieving competitive and in some cases even better results.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Gkiokas, Aggelos
- Γκιόκας, Άγγελος
Subjects
dc:subject × 21- Ανάκτηση μουσικής
- Ανάκτηση μουσικής πληροφορίας
- Μηχανική μάθηση
- Μηχανική μάθηση στην υπολογιστική μουσική
- Μουσικός ρυθμός
- Μουσική Ομοιότητα
- Μουσικό Τέμπο
- Νευρωνικά δίκτυα
- Music information retrieval
- Machine learning
- Music rhythm
- Music similarity measures
- TEMPO
- Artificial neural networks
- Beat Tracking
- Meter Estimation
- Restricted Boltzmann Machines
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38673