Parallel combination method

For more info about our services contact : help@bestpfe.com

Table of contents

1 Introduction 
1 Structure of the document
2 Outlines
2 Introduction- French Version 
1 Introduction
2 Résumé étendu par chapitre
3 Reminder of classification techniques involving combination of information 
1 Supervised pattern recognition
1.1 Features transformation and selection
1.2 Classifiers architecture
1.2.1 Generative approach: GMM
1.2.2 Discriminative approach: SVM
1.2.3 Instance-based approach: k-NN
1.3 Performance of classification
1.3.1 Measure of performance
1.4 Comparison of classification performances
2 Combination of information for classification problems
2.1 Levels of combination
2.2 Schemes for combination of classifiers decisions
2.2.1 Parallel combination
2.2.2 Sequential combination
2.3 Trainable .vs. non-trainable combiners
3 Conclusions
4 Singing voice: production, models and features 
1 Singing voice production
1.1 Vocal production
1.2 Some specificities of the singing production
1.2.1 Formants tuning
1.2.2 Singing formant
1.2.3 Vocal vibrato
1.2.4 Portamento and Legato
2 Models for voiced sounds
2.1 Source-filter model
2.1.1 Model description
2.1.2 Estimation of the vocal transfer function
2.2 Harmonic sinusoidal model
2.2.1 Model description
2.2.2 Sinusoidal model parameters estimation
2.3 Intonative model
2.3.1 Model description
2.3.2 Model parameters estimation
2.3.3 Model evaluation
2.4 Relation between intonative and source-filter model
2.4.1 Estimation of the formant position from a cycle of frequency modulation
3 Features for singing voice
3.1 Timbral features
3.2 Intonative features
4 Summary and Conclusions
5 Singing voice localization and tracking in polyphonic context 
1 Problems statement
1.1 Singing voice detection
1.2 Singing voice tracking
2 Related works
2.1 Singing voice localization
2.2 Instrument identification
2.2.1 Solo instrument identification
2.2.2 Multiple instruments recognition in polyphonic recordings .
2.3 Singing voice extraction using source separation approach
2.3.1 Blind Source Separation (BBS) approaches
2.3.2 Statistical Modeling approaches
2.3.3 Computational Auditory Scene Analysis (CASA) approaches
2.4 Singing melody transcription
3 Proposed approach for singing voice detection
3.1 Description of the proposed approach to localize vocal segments
3.1.1 Step 1 – Partial tracking and segmentation
3.1.2 Step 2 – Extraction of features
3.1.3 Step 3 – Selection of vocal partials
3.1.4 Step 4 – Formation of vocal segments
3.2 Evaluation
3.2.1 Data set
3.2.2 Results
4 Proposed approach to track harmonic contents of the singing voice
4.1 Description of the method to group harmonically related partials
4.1.1 Theoretical fundaments
4.1.2 Step 1 – Measure of similarity between partials
4.1.3 Step 2 – Distance matrix
4.1.4 Step 3 – Grouping harmonically related partials by clustering
4.2 Evaluation of clusters of harmonic partials
4.2.1 Data set
4.2.2 Measure for cluster evaluation
4.2.3 Results
4.3 Application to singing voice detection
4.3.1 Data set
4.3.2 Results
4.4 Application to multi-pitch and sung melody transcription
4.4.1 Method to estimate the f0 of a cluster
4.4.2 Data set
4.4.3 Measure
4.4.4 Results
5 Summary and conclusions
6 Singer Identification 
1 Problem statement
2 Related works
2.1 Artist identification
2.2 Singer identification
2.3 Perceptual recognition of singers
3 Proposed approach for singer identification
3.1 Description of the proposed approach based on the combination of timbral and intonative features
3.1.1 Sound descriptions complementarity
3.1.2 Combining decisions obtained with each sound description
3.2 Evaluation of the combination method
3.2.1 Data sets
3.2.2 Results
4 Proposed approach to evaluate the features robustness for singer recognition
4.1 Description of the proposed approach
4.2 Evaluation of the features of robustness against intra-song and inter-song variations
4.2.1 Data set
4.2.2 Results for intra-song variations on a cappella recordings
4.2.3 Results for inter-song variations
5 Summary and conclusions
7 Conclusions 
1 Summary
2 Future works