haku: @instructor Miche, Yoan / yhteensä: 6
viite: 5 / 6
Tekijä: | Ramaseshan, Ajay |
Työn nimi: | Application of multiway methods for dimensinality reduction to music |
Julkaisutyyppi: | Diplomityö |
Julkaisuvuosi: | 2013 |
Sivut: | 86 s. + liitt. 5 Kieli: eng |
Koulu/Laitos/Osasto: | Perustieteiden korkeakoulu |
Oppiaine: | Tietokoneverkot (T-110) |
Valvoja: | Simula, Olli |
Ohjaaja: | Corona, Francesco ; Miche, Yoan |
Elektroninen julkaisu: | http://urn.fi/URN:NBN:fi:aalto-201407052298 |
OEVS: | Sähköinen arkistokappale on luettavissa Aalto Thesis Databasen kautta.
Ohje Digitaalisten opinnäytteiden lukeminen Aalto-yliopiston Harald Herlin -oppimiskeskuksen suljetussa verkossaOppimiskeskuksen suljetussa verkossa voi lukea sellaisia digitaalisia ja digitoituja opinnäytteitä, joille ei ole saatu julkaisulupaa avoimessa verkossa. Oppimiskeskuksen yhteystiedot ja aukioloajat: https://learningcentre.aalto.fi/fi/harald-herlin-oppimiskeskus/ Opinnäytteitä voi lukea Oppimiskeskuksen asiakaskoneilla, joita löytyy kaikista kerroksista.
Kirjautuminen asiakaskoneille
Opinnäytteen avaaminen
Opinnäytteen lukeminen
Opinnäytteen tulostus
|
Sijainti: | P1 Ark Aalto 8728 | Arkisto |
Avainsanat: | Mel spectrogram MDS MLSCA MPCA music collection MIR PCA |
Tiivistelmä (eng): | This thesis can be placed in the broader field of Music Information Retrieval (MIR). MIR refers to a huge set of strategies, software and tools through which computers can analyse and predict interesting patterns from audio data. It is a diverse and multidisciplinary field, encompassing fields like signal processing, machine learning, and musicology and music theory, to name a few. Methods of dimensionality reduction are widely used in data mining and machine learning. These help in reducing the complexity of the classification/clustering algorithms etc., used to process the data. They also help in studying some useful statistical properties of the dataset. In this Master's Thesis, a personalized music collection is taken and audio features are extracted from the songs, by using the Mel spectrogram. A music tensor is built from these features. Then, two approaches to unfold the tensor and convert it into a 2-way data matrix are studied. After unfolding the tensor, dimensionality reduction techniques like Principal Components Analysis (PCA) and classic metric Multidimensional Scaling (MDS) are applied. Unfolding the tensor and performing either MDS or PCA is equivalent to performing Multiway Principal Component Analysis (MPCA). A third method Multilevel Simultaneous Component Analysis (MLSCA), which builds a composite model for each song is also applied. The number of components to retain is obtained by hold-out validation. The fitness of each of these models were evaluated with the T2 and Q statistic, and compared with each other. The aim of this thesis is to produce a dimensionality reduction which can be used for further MIR tasks like better clustering of data with respect to e.g. artists / genres. |
ED: | 2014-01-07 |
INSSI tietueen numero: 48301
+ lisää koriin
INSSI