Learning to analyse tonal music Pl á cido Rom á n Illescas David Rizo Jos é Manuel I ñ esta Pattern recognition and Artificial Intelligence group University.

Slides:



Advertisements
Similar presentations
AP Music Theory Ms. Friedrichs.
Advertisements

Music Analysis and Generation Supervisors: Jon McCormack & Lloyd Allison Final Presentation Oliver Ng (B.DigSys)
Music Analysis and Generation Supervisors: Jon McCormack & Lloyd Allison Interim Presentation Oliver Ng (B.DigSys)
Tennessee High School Survey Summary Graphs Percentage of students who: Note: This graph contains weighted results. See the corresponding summary tables.
Speaker Associate Professor Ning-Han Liu. What’s MIR  Music information retrieval (MIR) is the interdisciplinary science of retrieving information from.
Honolulu, 23 rd of May 2011PESOS Evaluating the Compatibility of Conversational Service Interactions Sam Guinea and Paola Spoletini.
Algorithms + L. Grewe.
Chapter 2 – Scales, Tonality, Key, Modes
Outline Introduction Music Information Retrieval Classification Process Steps Pitch Histograms Multiple Pitch Detection Algorithm Musical Genre Classification.
Pitch-spelling algorithms David Meredith Aalborg University.
Melodic Organization Chapter 6. Motive Short melodic and/or rhythmic pattern Usually only a few beats Recurs throughout a piece or section Unifying element.
Combining Inductive and Analytical Learning Ch 12. in Machine Learning Tom M. Mitchell 고려대학교 자연어처리 연구실 한 경 수
Automated Fugue Generation Yu Yue Yue Yang supervised by Prof Andrew Horner.
What is music? Music is the deliberate organization of sounds by people for other people to hear.
Artificial Neural Networks ML Paul Scheible.
Tree structured representation of music for polyphonic music information retrieval David Rizo Departament of Software and Computing Systems University.
Learning with Bayesian Networks David Heckerman Presented by Colin Rickert.
Localized Key-Finding: Algorithms and Applications Ilya Shmulevich, Olli Yli-Harja Tampere University of Technology Tampere, Finland October 4, 1999.
1 MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING By Kaan Tariman M.S. in Computer Science CSCI 8810 Course Project.
Comparison of Scale Degrees In Major and minor. 1. Scale Degrees ^1, ^2, ^4, and ^5 = Tonal Scale Degrees same for major and all three forms of the minor.
Testing an individual module
Voice Separation A Local Optimization Approach Voice Separation A Local Optimization Approach Jurgen Kilian Holger H. Hoos Xiaodan Wu Feb
Advanced Algorithm Design and Analysis Student: Gertruda Grolinger Supervisor: Prof. Jeff Edmonds CSE 4080 Computer Science Project.
Music Composition by Dr. Lai Sheung Ping 31st January, 2007.
Thursday, October 18,  Music Sharing!  Review: Common Harmonies (major & minor)  Review: Cadences  New: Passing Tones & Neighboring Tones 
Chapter 11: Artificial Intelligence
HANA HARRISON CSE 435 NOVEMBER 19, 2012 Music Composition.
Chapter 14: Artificial Intelligence Invitation to Computer Science, C++ Version, Third Edition.
David Temperley Presentation by Carley Tanoue
Voice Leading in Four-Part Chorale Writing
CMSC 345 Fall 2000 Unit Testing. The testing process.
M ULTIFRAME P OINT C ORRESPONDENCE By Naseem Mahajna & Muhammad Zoabi.
ANTs PI Meeting, Nov. 29, 2000W. Zhang, Washington University1 Flexible Methods for Multi-agent distributed resource Allocation by Exploiting Phase Transitions.
Kansas State University Department of Computing and Information Sciences CIS 830: Advanced Topics in Artificial Intelligence Friday, February 4, 2000 Lijun.
RoBach A Case Study in the Use of Genetic Algorithms for Automatic Music Composition.
Rhythmic Transcription of MIDI Signals Carmine Casciato MUMT 611 Thursday, February 10, 2005.
A year 1 musicianA year 2 musicianA year 3 musician I can use my voice to speak, sing and chant. I can use instruments to perform. I can clap short rhythmic.
Bernd Möbius CoE MMCI Saarland University Lecture 7 8 Dec 2010 Unit Selection Synthesis B Möbius Unit selection synthesis Text-to-Speech Synthesis.
Multi Way Selection You can choose statement(s) to run from many sets of choices. There are two cases for this: (a)Multi way selection by nested IF structure.
Polyphonic Transcription Bruno Angeles McGill University - Schulich School of Music MUMT-621 Fall /14.
Lexical Analysis: Finite Automata CS 471 September 5, 2007.
A n = c 1 a n-1 + c2an-2 + … + c d a n-d d= degree and t= the number of training data (notes) The assumption is that the notes in the piece are generated.
Music Genre Classification Alex Stabile. Example File
MELODIC WRITING. FINISH THIS MELODY! TO BREAK RULES, YOU HAVE TO LEARN THEM FIRST… Composers often consider many things when they write music- rhythm,
 Based on observed functioning of human brain.  (Artificial Neural Networks (ANN)  Our view of neural networks is very simplistic.  We view a neural.
Key Relationships AP MUSIC THEORY 2014.
West Virginia Middle School Survey Summary Graphs Percentage of students who: Note: This graph contains weighted results. See the corresponding summary.
Concepts and Realization of a Diagram Editor Generator Based on Hypergraph Transformation Author: Mark Minas Presenter: Song Gu.
 6 th Musical Literacy 1.1 All students will be able to use a steady tone when performing.
A framework for answering aural questions using the 6 concepts of music.
Write a function rule for a graph EXAMPLE 3 Write a rule for the function represented by the graph. Identify the domain and the range of the function.
Alex Stabile. Research Questions: Could a computer learn to distinguish between different composers? Why does music by different composers even sound.
Computer Aided Composition Kevin Wampler. Assisted Notation and Layout Automated Composition Style-driven Suggestions Alternative Notations Automatic.
Melody Recognition with Learned Edit Distances Amaury Habrard Laboratoire d’Informatique Fondamentale CNRS Université Aix-Marseille José Manuel Iñesta,
Neural networks (2) Reminder Avoiding overfitting Deep neural network Brief summary of supervised learning methods.
Melody Characterization by a Fuzzy Rule System Pedro J. Ponce de León, David Rizo, José M. Iñesta (DLSI, Univ. Alicante) Rafael Ramírez (MTG, Univ. Pompeu.
Stochastic Text Models for Music Categorization Carlos Pérez-Sancho, José M. Iñesta, David Rizo Pattern Recognition and Artificial Intelligence group Department.
HAYDN 4 Symphony No. 26 in D minor, Lamentatione: movement i.
Network Management Lecture 13. MACHINE LEARNING TECHNIQUES 2 Dr. Atiq Ahmed Université de Balouchistan.
A shallow description framework for musical style recognition Pedro J. Ponce de León, Carlos Pérez-Sancho and José Manuel Iñesta Departamento de Lenguajes.
Genre Classification of Music by Tonal Harmony Carlos Pérez-Sancho, David Rizo Departamento de Lenguajes y Sistemas Informáticos, Universidad de Alicante,
Pattern Recognition Lecture 20: Neural Networks 3 Dr. Richard Spillman Pacific Lutheran University.
Fanfare 3 Year 10 Composition.
Rule Induction for Classification Using
AP Music Theory Mr. Silvagni
Integrating Segmentation and Similarity in Melodic Analysis
Test Case Test case Describes an input Description and an expected output Description. Test case ID Section 1: Before execution Section 2: After execution.
MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING
MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING
Chord Recognition with Application in Melodic Similarity
Presentation transcript:

Learning to analyse tonal music Pl á cido Rom á n Illescas David Rizo Jos é Manuel I ñ esta Pattern recognition and Artificial Intelligence group University of Alicante Helsinki; 9th July, 2008

MML’08 Tonal analysis Musical analysis is a mean to better understand the thought of the composer when creating a piece Key: Bm composed How should it be played? Understanding it through analysis Sequence of notes  sequence of chords Chord has a Tonal Function: { T | D | SD } Equivalent in NLP to Syntax analysis Tonal function = {tension | relaxation}

MML’08Objectives To analyze a musical piece To get good correct analysis percentages describing the solution in a human-readable way To integrate harmonic analysis in other MIR and e-learning systems Automatic composition Expressive automatic performance Score reduction Pitch spelling Harmonic comparison of works Applications

MML’08 State of the art GrammarsWinograd 1968 / 1992: no melodic analysis Chemilier 2004, Steedman 1984: jazz chord progressions Expert systemsMaxwell 1992: problem with arpeggiated chords Probabilistic modelsRaphael 2004: no human readable explanation of solution Preference rulesTemperley 1999: errors in analysis Scoring techniquesPardo 2002: no tonal functions Neural networksHornel 1996: no human readable explanation of solution Model matchingTaube 1999: discontinued

MML’08Methodology

1st: melodic analysis Target: get notes tagged as HT (harmonic tone) or NHT (non- harmonic tone) Rules system based on music theory Rules summary

MML’08 2nd: Key filtering w0w0 w1w1 w2w2 w3w3 w4w4 w5w5 w6w6 w7w7 w8w8 Split into time frames Remove the keys that do not match an expected semitones table Semitone differences from the tonic of the specified degree (the (1) value represents the Neapolitan, the (4) represents the Picardy ending)

MML’08 3rd. Weighted acyclic directed graph 3.1: Chord extraction Extract all possible chords for each time frame. Use backtracking to get all valid succesions of thirds 3 3

MML’08 3rd. Weighted acyclic directed graph

MML’08 3rd. Weighted acyclic directed graph etc. Extract of the actual graph 3.2: Construction of weighted acyclic directed graph Nodes are valid combinations of key+chord+tonal function for a time frame

MML’08 Edges correspond to cadences. Dashed lines represent invalid cadences. Weights are set according to the progression 3rd. Weighted acyclic directed graph Dynamic programming: get best path

MML’08Weights System performance is determined by a correct weight tuning Two approaches: – To use domain knowledge to set the weights As reported in a previous work (ICMC’07) – To learn the weights using a machine learning algorithm: genetic algorithm approach The approach presented here (MML’08)

MML’08 Genetic algorithm Chromosome: 1 gene for each weight – 25 genes, integer range [  3000, 1000] Fitness function – minimize number of incorrectly analyzed windows – ground-truth of 6 Bach chorales (BWV-253, 26, 437, 29, 272, and 438) Total: 832 time windows 300 iterations Iterations Fitness

MML’08 Output of trained system

MML’08Experiments Domain knowledge weights vs. Learnt weights – Tonal Function (TF) and Tonality (T) success rates Without GAWith GA TF (%)T (%)TF (%)T (%) BWV BWV BWV BWV BWV BWV Better TF than T suggests errors in tonality mode

MML’08Conclusion The system with weights learnt from data – avoids the arbitrariness or subjectivity of human expert – could learn different models for different styles Future works: – To build a larger tagged corpus to learn graph weights – To learn the model using examples from other genres – To construct chords from sequences of notes to cope with arpeggiated chords and monodies