5
M LP LY R IC AL SYNTH ESIS FO R PREDIC TO R M U SIC AL EXPRESSIO NS 2003 – 12 – 18 ECE 539 Introduction to : A ritificialN eural N etw ork and Fuzzy System s KojiYabum oto 9017180622

MLP Lyrical Analysis

  • Upload
    hugh

  • View
    19

  • Download
    0

Embed Size (px)

DESCRIPTION

MLP Lyrical Analysis. % of Unique Words # of Unique Words Average Word Length # of Lyrics # of Characters. Input Feature Vectors:. C Application. Traversal of directory in search of lyric data (*.lyr) Parsing and loading lyrics into proper data array structure. - PowerPoint PPT Presentation

Citation preview

Page 1: MLP Lyrical Analysis

MLP LYRICAL SYNTHESIS FOR PREDICTOR

MUSICAL EXPRESSIONS

2003 – 12 – 18

ECE 539 Introduction to :

Aritificial Neural Network

and Fuzzy Systems

Koji Yabumoto9017180622

Page 2: MLP Lyrical Analysis

MLP Lyrical Analysis

● % of Unique Words● # of Unique Words● Average Word Length● # of Lyrics● # of Characters

Input Feature Vectors:

Page 3: MLP Lyrical Analysis

C Application

● Traversal of directory in search of lyric data (*.lyr)

● Parsing and loading lyrics into proper data array structure.

● Filtering of data skewing characters.● Analysis to extract needed characteristics of

lyrics● Output into file with proper format for MLP

program.

Page 4: MLP Lyrical Analysis

MLP Development● Normalization of

Feature Vectors ● Optimal solution for #

of layers and # of neurons/layer.

● Compete Against Baseline Kmeans algorithm (~70%) Rate

● Try to achieve a Test Crate nearly as good as Train Crate

2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

100

Crate Test,Train 2-Class MLP

Train Crate

Test Crate

Number of hidden layers

Cla

ssifi

catio

n (

%)

Page 5: MLP Lyrical Analysis

Modifications to Original Specification

● Study of data input feature vectors to determine correlation with classification.

● Changing the size of the ouput classification to improve performance.

● Study of different types of data's effectiveness.