Στην παρούσα διπλωματική εργασία γίνεται μελέτη των στατιστικών παραμέτρων ηχητικών σημάτων. Μελετάται η δυνατότητα αυτόματης ταξινόμησης μουσικής ανά είδος, η οποία βασίζεται στην εξαγωγή αυτών των παραμέτρων. Επίσης, μελετάται η μεταβολή αυτών σε συνθήκες αντήχησης, δίνοντας έμφαση στην παράμετρο φασματικής ασυμμετρίας ηχητικού σήματος. Σε αυτό το πλαίσιο, προτείνεται μέθοδος κατασκευής μοντέλου πρόβλεψης της συμπεριφοράς της συγκεκριμένης παραμέτρου σε συνθήκες αντήχησης, που στόχο έχει την εκτίμηση της απόστασης ηχητικής πηγής – δέκτη σε κλειστό χώρο, καθώς και την πρόβλεψη της ανωτέρω παραμέτρου ανηχωικού σήματος από σήματα με αντήχηση. / In this thesis we study the audio features and their applications, such as automatic music genre classification. It is also studied the behavior of these features under reverberant conditions, emphasizing on spectral skewness. In this framework, it is suggested a method of predicting the behavior of this feature under reverberant conditions, which could have many applications such as source - receiver distance estimation and prediction of the spectral skewness of anechoic audio signals.

Dans cet ouvrage, nous définissons précisément ce qu’est l’identification audio à grande échelle. En particulier, nous faisons une distinction entre l’identification exacte, destinée à rapprocher deux extraits sonores provenant d’un même enregistrement, et l’identification approchée, qui gère également la similarité musicale entre les signaux. A la lumière de ces définitions, nous concevons et examinons plusieurs modèles d’empreinte audio et évaluons leurs performances, tant en identification exacte qu’en identificationapprochée. / N this work we give a precise definition of large scale audio identification. In particular, we make a distinction between exact and approximate matching. In the first case, the goal is to match two signals coming from one same recording with different post-processings. In the second case, the goal is to match two signals that are musically similar. In light of these definitions, we conceive and evaluate different audio-fingerprint models.

This diploma work deals with music genre recognition using the techniques of Music Information Retrieval. It contains a brief description of the principle of this research area and its subfield called Music Genre Recognition. The following chapter includes selection of the most suitable parameters for describing music genres. This work further characterizes machine learning methods used in this field of research. The next chapter deals with the descriptions of music datasets created for genre classification studies. Subsequently, there is a draft and evaluation of the system for music genre recognition. The last part of this work describes the results of partial parameter analysis, dependence of genre classification accuracy on the amount of parameters and contains a discussion on the causes of classification accurancy for the individual genres.

Music listening is considered one of the most ubiquitous activities in everyday life, and one of the main reasons why people listen is to affect and regulate their mood. The vast availability and unlimited access of music has made it difficult to find relevant music that fits both the context and the preferences of the music listener. The aim of this project was to investigate the personalized relationship between music and mood using everyday technologies, focusing on how a listening experience could be adapted to the desired affect of a music listener while also taking the user’s individual listening history into account. In large, the project concentrated on the possibility of using context-aware music recommendation to generate personalized playlists by focusing on the audio features and corresponding mood of the music. A web-based application was developed to act as a prototype for the study, where the application allowed users to connect to Spotify, pick a desired mood and generate a playlist. By allowing people to access music in this personalized way, a user study could be conducted in order to investigate their music listening while incorporating this recommendation tool. The findings showed that the users’ found the experience to be engaging in that they could use the application as a companion to everyday tasks in addition to it being a tool for getting new, personalized music recommendations. Overall, the participants also found the generated playlists to be accurate to their music preferences and desired affective state.

Music mood recognition by machine continues to attract attention from both academia and industry. This thesis explores the hypothesis that the music emotion problem is circular, and is a primary step in determining the efficacy of circular regression as a machine learning method for automatic music mood recognition. This hypothesis is tested through experiments conducted using instances of the two commonly accepted models of affect used in machine learning (categorical and two-dimensional), as well as on an original circular model proposed by the author. Polygonal approximations of circular regression are proposed as a practical way to investigate whether the circularity of the annotations can be exploited. An original dataset assembled and annotated for the models is also presented. Next, the architecture and implementation choices of all three models are given, with an emphasis on the new polygonal approximations of circular regression. Experiments with different polygons demonstrate consistent and in some cases significant improvements over the categorical model on a dataset containing ambiguous extracts (ones for which the human annotators did not fully agree upon). Through a comprehensive analysis of the results, errors and inconsistencies observed, evidence is provided that mood recognition can be improved if approached as a circular problem. Finally, a proposed multi-tagging strategy based on the circular predictions is put forward as a pragmatic method to automatically annotate music based on the circular model. / Graduate / 0984 / 0800 / 0413 / zazz101@hotmail.com