Basic Spectrogram
Lab 8
Spectrograms
 Spectrograph: Produces visible patterns of
acoustic energy called spectrograms
 Spectrographic Analysis:

Acoustic signal changes rapidly & continuously

Need a dynamic analysis revealing spectral
features

Short term running spectrum
Spectrographic Analysis
 Time, Frequency & intensity

Time (horizontal axis): Phonetic elements
will be from left to right

Frequency (Vertical axis): Increasing in
upward direction

Intensity represented by a gray scale or as
variations in darkness
Dimensions of Spectrograms
A. Burst of noise
B. Vowel with 4 formants
Time
C. Noise with high frequency energy
Source/Filter
 Resonator & Source Energy

Filter = Vocal tract
• Energy passed in a frequency selective manner
• Production of different vowels changes the filter shape

Source = Harmonic spectrum of voicing
• Fundamental frequency & its harmonics
• Source spectrum is the acoustic energy activating the formants
 Source-filter theory of vowel production:

The energy form the source (vibrating vocal folds) is
modified by the resonance characteristics of the filter
(vocal tract)
Formants- Filter
 Are formed through resonation of sound in
the vocal tract (filter function)
 Can only see through a wide band
spectrogram
F2
F2
F1
Harmonics- Source
Harmonics (Spectrum)
Harmonics(Spectrogram)
Source-Filter Theory
Source-Filter
Wide Band vs. Narrow Band
 Wide Band: analyzing filter wide, 300400 Hz

Good to show formants because they
have a wide spread of acoustic energy
 Narrow Band: analyzing filter is of
higher resolution, 100 Hz

Good to show the harmonics of source
spectrum
Bandwidth
Narrow
Bandwidth
(Harmonics)
Wide
Bandwidth
(Formants)
Continuous Spectrum Bandwidth
Vowel: Wide Band vs. Narrow Band
Wide Band Spectrogram
Fricative noise:
Above 3 or 4 kHz
Diphthong: Horizontal
bands of energy
Digital Signal Processing
 Basic objective of digital signal processing:


Convert the analog acoustic signal (Panel A) to a digital
form (series of #’s)
How do we get a waveform into a digital computer?
• Analog-Digital conversion (A-D): results in samples of time
and amplitude

Correct sampling rate is important to reconstruct the
wave form
• Sampling rate must be 2x the bandwidth of analysis: Nyquist
Frequency
– Ex. Sample speech with a bandwidth of 5000 Hz, than the
sampling rate should be 10,000 Hz
Analog-Digital Conversion
b. Time sampling intervals
Laboratory
 Part I: Recording the sample- Sampling rate


Set sampling rate for shortest duration possible
Sustain /a/: **open 3 views to print all samples on 1 screen
• Sample at 20,000 samples/sec; look at 5-6 cycles
• Sample at 5,000 samples/sec; look at 5-6 cycles
• Sample at 2,500 samples/sec; look at 5-6 cycles


What happens as sampling rate drops
Narrow band filter each sample (print each)
Laboratory
 Part II: Filter bandwidth

Record “I am tall”
• Wide band and narrow band the statement
– Find the following characteristics on one or both of the
spectrograms:
» Harmonic lines
» Formant bands
» Vertical striations
» Acoustic indication of release of air pressure for /t/
– Which has better frequency resolution? Time resolution?
Which can you determine formant bands?
Laboratory
 Part II: Cont.

Make wide and narrow band spectrograms of
the following, stressing the underlined word:
•
•
•
•

“I am tall”
“I am tall”
“I am tall”
“I am tall?”
Which spectrograms show the intonation
contour? Mark the contour on one of your
printouts
Laboratory
 Part III: Changing the Filter

Say /i:u:i/ without changing intonation
• Prepare 2 spectrograms of this utterance
– 1st- 2x’s your fundamental frequency
– 2nd- no more than half your fundamental
• Answer questions:
– Label each vowel on each spectrogram
– Harmonics
– Formants