Metode razlikovanja govora i glazbe u digitalnim zvučnim zapisima

master's thesis

• Title (english): Methods for speech and music discrimination in digital soundtracks
• Author: Ivan Križanić
• Mentor: Davor Petrinović (mentor)
• Committee member: Davor Petrinović (predsjednik povjerenstva)
• Committee member: Damir Seršić (član povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2015-07-10, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

U sklopu ovog diplomskog rada obrađen je povijesni pregled na metode razlikovanja govora i glazba te je napravljen pregled različitih čimbenika koji utječu na uspješnost klasifikacije. To su prije svega parametri zvuka, u vremenskoj domeni postotak okvira niske energije i broj prolaza kroz ništicu, u frekvencijskoj domeni spektralni centroid, spektralni tok, frekvencija većinske spektralne snage te u kepstralnoj domeni modulacijska energija na 4 Hz i modul razlike spektra i rekonstruiranog spektra iz kepstra. Rad se zatim fokusira na različite klasifikatore koje je moguće koristiti u sklopu razlikovanja govora i glazbe te obrađuje četiri različita klasifikatora. Na kraju se diskutiraju dvije različite gotove metode koje se mogu koristiti za klasificiranje govora i glazbe.

Abstract (english)

In the first part of this Masters' Thesis there is a historical overview of the methods for Speech/Music Discrimination and a survey of various factors that affect the classification performance. These factors consist primarily of parameters of sound. In the time domain, they include percentage of low-energy frames and zero crossing rate, in the frequency domain there are spectral centroid, spectral flux, spectral rolloff point and in the cepstral domain there are 4 Hz modulation energy. The Thesis then focuses on different classifiers that can be used for speech/music discrimination, pinpointing four different classifiers. Finally, two different methods that can be used to classify speech and music are discussed.

• Keywords: digitalna obrada zvuka; razlikovanje govora i glazbe; broj prolaza kroz ništicu; spektralni tok; spektralni centroid; MFCC koeficijenti; MAP estimator; Gaussove mješavine; k-d stabla; k-NN algoritam
• Keywords (english): digital sound processing; speech/music discrimination; zero-crossing rate; spectral flux; spectral centroid; MFCC coefficients; the MAP estimator; Gaussian mixtures; k-d trees; k-NN algorithm
• Language: croatian
• URN:NBN: urn:nbn:hr:168:931134
• Study programme: Title: Electrical Engineering and Information Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije (magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 09:56:35