Spectral Parameters of Normal

1
Spectral Parameters of Normal Abnormal Speech

• Rick McGuire
• Saint Louis University

2
Overview

• WaveSurfer was used in examples and is free and
  available on the internet (www.speech.kth.se/wav
  esurfer/)
• Parameters specific quantities that represent
  an aspect of speech
• We will review
• Waveforms, spectral analyses spectrograms
• Parameter analysis comparisons within speaker
  between-speaker
• Parameters associated with
• Simplification of syllables
• Vowel, glide and liquid variants
• Analysis of normal abnormal stops
• Fricative analysis
• Never loose site of the direct correspondence
  between articulatory physiology speech
  acoustics.

3
Spectrograms

• Three-dimensional display of time (horizontal
  axis), frequency (vertical axis), intensity
  (gray scale or color)
• Narrow-band (spectrogram) analysis
• analysis filter adjusted narrow (such as, 45 hz)
  resulting in increased frequency resolution
• Good for examining fundamental frequency
  harmonics of voice
• Wide-band (spectrogram) analysis
• Relatively wide analysis filter setting (such as,
  256 Hz) resulting in greater time resolution
• Good for examining formant frequency patterns

4
(No Transcript)
5
Spectral Analysis

• Spectrum display showing distribution of
  intensity (energy) as a function of frequency
• Ways to obtain spectra
• Fourier transform conversion of values in time
  domain (waveform) into values in frequency domain
  (spectrum) (e.g., FFT, DFT)
• Linear predictive coding (LPC) uses a weighted
  linear sum of samples to predict upcoming values
  resulting in a spectrum

6
(No Transcript)
7
Spectral Parameter Analysis

• Within-Speaker Between-Speaker Comparisons
• Duration measurements (temporal parameters)
• Spectral descriptions/measurements (e.g.,
  spectral energy, energy variability, formants)
• Perceptual descriptions the ear remains your
  best tool for these comparisons
• Visual and linguistic cues are also critical in
  making comparisons

8
Within-Speaker Comparison ????-????
4 year old female
devoiced
s
?
?
?
?
p
m
9
(No Transcript)
10
Simplification of Syllable Structure

• Cluster Reduction
• Final Consonant Deletion

4 year old female
Target star
?
?
Room noise
11
/r/ Distortion
6 year old male
?
?
?
?
?
??
Target Spencer
12
(lisp)
Lateralized ????
4 year old female
13
Abnormal Stop prevocalic Voicing
4 year old male
Final stop consonant deletion in both syllables
Target quack quack
?
?
?
?
?
?
?
14
Summary

• The cost of acoustic analysis systems has
  prohibited clinicians from embracing spectral
  analysis cost is no longer a factor for this
  type of application
• Acoustic/spectral analyses are good supplements
  to traditional perceptual analyses
• In some cases, acoustic data my change perceptual
  judgments
• Acoustic/spectral analyses can easily be extended
  into remediation as a metric of progress

15
Summary

• Clinical applications such as, VideoVoice
  SpeechViewer use automated spectral matching
• Moving the monkey up the tree relates to spectral
  matching to a target (spectral) model
• Spectral parameters are the result of speech
  physiology respiration, phonation, resonance,
  and articulation
• Therefore, improvement in spectral matching is
  a product of the place, manner, voicing, and
  timing of the utterance.
• Louder and higher have little effect on
  spectral signatures