Neural Net Algorithms for SC Vowel Recognition

1
Neural Net Algorithms for SC Vowel Recognition

• Presentation for EE645
• Neural Networks and Learning Algorithms
• Spring 2003
• Diana Stojanovic

2
Summary

• Neural net algorithms applied to recognition of
  Serbo-Croatian vowels
• Follows Thubthong Kijsirkul (2001) paper on
  Thai phoneme recognition
• Light background will be provided

3
Introduction

• Speech recognition has many applications (PCs,
  cell phones, home appliance activation a la
  Dilbert etc.)

4
Introduction 2

• There are various algorithms for recognizing
  speech, some of which rely on the recognition of
  individual phonemes or sounds

5
Block diagram of speech recognition system

• For this project
• Signal Processing segmentation, spectral
  analysis
• Speech Recognition Individual vowel recognition

Signal Processing
Speech Recognition
6
Previous work

• Thubthong Kijsirkul (2001) tested multi-class
  Support Vector Machine (SVM) vs. Multilayer
  Perceptron (MLP) for recognition of Thai Vowels
  and tones
• They claim superiority of SVM, while the
  recognition rate differs by 2-3 for comparably
  complex systems

7
About speech sounds

• Speech sound is an acoustic wave
• Speakers vocal tract shapes the spectrum of each
  sound
• Spectrum depends on the speaker and on the
  property of the particular sound (for instance
  /u/), thus recognition in spectral domain is
  possible

8
Vowel Formants

• Vowels can be recognized in spectral domain by
  the characteristic lines corresponding to their
  properties (backness, height, lip rounding etc.)
• These lines formants- occur at resonant
  frequencies of the vocal tract

9
Serbo-Croatian Vowel Chart
10
Data Used in the Project

• Data collection and Properties
• Type of speech speaker dependent, accented
  syllables
• 480 isolated words were recorded and digitized at
  11 kHz
• Vowels in accented position segmented manually
• Vowel formants measured by PCQuirer

11
Sound Features Measured

• Only first two formants were used for training
  the nets in order to reduce complexity
• Based on the property of the SC sounds, the
  performance should not suffer from this low
  dimensionality

12
Perceptron,Backprop and Support Vector Machine

• We learned about this throughout the semester ?.
• For details, please refer to the paper

13
Results
14
What is next?

• First, finish the SVM results
• Examine fast, connected speech
• Speaker independent recognition