Speech Acoustic Measurement & Analysis

Question 1

Sound Spectrograph

Answer 1

Developed during WWII during the process of trying seeking methods to encode and decode messages
Able to display formants as a continuous function of time

Question 2

Spectrum Analysis

The spectrum analyzer performs its analysis

Answer 2

The spectrum analyzer performs its analysis by moving a fixed-width analysis band, or filter, across the entire frequency range.
▪
If the analysis band is swept continuously across the entire frequency range, it will provide overall voltage outputs as a continuous

Question 3

Schematic diagram showing how the spectrograph

Answer 3

Schematic diagram showing how the spectrograph performs spectral analysis by sweeping an analysis band of fixed width (e.g., 300 Hz) across the frequency range of interest, and recording the average voltages from the analysis band as a continuous

Question 4

The Original Sound Spectrograph

The invention of the spectrograph allowed for the study

Answer 4

The invention of the spectrograph allowed for the study of the time-varying acoustic results of articulatory processes
▪ If articulator movements are changing as a function of time, and therefore changing vocal tract configuration as a function of time, the changes are reflected in formant transitions—formant frequencies that change over time

Question 5

The Original Sound Spectrograph

The time-varying patterns of electromagnetic strength are

Answer 5

The time-varying patterns of electromagnetic strength are submitted to a spectrum analyzer in the form of time- varying voltages, where voltage is proportional to sound intensity (greater voltage = greater intensity) and the speed with which the voltage changes is proportional to frequency (faster voltage changes [shorter periods] = higher

Question 6

The Original Sound Spectrograph

The energy in the spectrum is sampled using an analysis band

Answer 6

The energy in the spectrum is sampled using an analysis band, or filter, that has a bandwidth of 300 Hz that is swept continuously across the entire frequency range of interest.
Because the voltage output from the analysis band is available for all frequencies and at every point in time, the spectrograph creates a total picture of the

Question 7

Digital Spectrograms

Produced almost instantaneously after

Answer 7

Produced almost instantaneously after an utterance has been recorded.
▪ The principles previously discussed for the original spectrograph are basically the same in digital spectrograms; the frequency and amplitude analysis are performed by moving a digital filter from low to high frequencies
▪
The output is a digital magnitude that is proportional to amplitude as a function of frequency.

Question 8

Interpretation of Spectrograms

Important features

Answer 8

Important features of the spectrographic display include:
the x-, y-, and z-axes
glottal pulses
formant frequencies
silent intervals
stop bursts aperiodic intervals.

The spectrogram shows a series of chunks, or segments, as the pattern is inspected from left to right.
The chunks, or segments, are important because they often correspond roughly to speech sounds.

Question 9

Axes

Answer 9

X-axis: time
Y-axis: frequency
Z-axis: intensity (third dimension of the spectrogram)
coded by the darkness of the pattern at any point on the spectrographic display

Question 10

Glottal Pulses

Dark vertical lines that are

Answer 10

Dark vertical lines that are an acoustic result of vocal fold vibration
Each individual line reflects a single glottal pulse–a point of excitation, when the vocal folds close quickly at the end of a glottal cycle and create a pressure wave whose spectrum is shaped by the vocal tract filter.

Question 11

Formant Frequencies

The dark bands seen in the patterns

Answer 11

The dark bands seen in the patterns with regularly spaced glottal pulses This pattern is seen for any speech sound produced with a relatively open vocal tract and voicing, including vowels, diphthongs, and semivowels (/l/, /w/, /ɹ/, /j/)
▪
Nasals are voiced and radiate sound

Question 12

Formant Frequencies

The formant frequencies change as

Answer 12

The formant frequencies change as a function of time in connected speech. The constant movement of the formants during speech production reflects the constant change in the configuration of the vocal tract.

Question 13

Silent Intervals and Stop Bursts

Appears as a

Answer 13

Appears as a brief blank spot, or gap on
the spectrogram
▫ Because the vocal tract is, in theory,
completely sealed during the closure, acoustic energy should not be radiated from the vocal tract
▪
If intensity is scaled on a spectrogram as the darkness of the trace, a white or nearly white segment indicates a

Question 14

Silent Intervals and Stop Bursts

For voiced stops, there is a small amount of

Answer 14

For voiced stops, there is a small amount of periodic energy on the baseline due to vocal fold vibration during vocal tract closure
▪ Vocal fold vibration during a closure interval causes vibration of the walls of the vocal tract,
▪ This energy is seen only in the lowest freqs of the spectrogram because walls of the vocal tract vibrate only at the lowest freqs of vocal fold vibration, filtering out the higher source harmonics

Question 15

Pauses vs Stop Closure

Answer 15

Pauses in speech are typically at least 150 ms and voiceless stop closure intervals are typically less than 120 ms.
▪ Many scientists have adopted a criterion of 200 ms–Silent intervals 200 ms or greater are identified as pauses, those less than 200 ms are subject to further evaluation
▫ This criterion is less reliable in certain speech disorders

Question 16

Aperiodic Intervals

Answer 16

Shown in a spectrogram as an interval of energy having no repeating pattern
These aperiodic intervals are most commonly associated with fricatives and the release phase of stops and affricates. Aperiodic energy may also be mixed with periodic energy for certain sound segments (such as voiced fricatives) or phonation types (such as breathy voice).

Question 17

Segmentation of Spectrograms

Answer 17

Identification of pieces of the display that correspond roughly to phonemic or phonetic units
Glottal pulses play an important role in the measurement of various attributes of the spectrogram
▪
▫
Often used to identify onsets and offsets of segments in a spectrogram

Question 18

Segmentation of Vowels

Answer 18

Segmentation of vowels is fairly simple when they are located between two obstruents
▪ The beginning of the vowel is taken as the first “full” glottal pulse, and the end of the vowel is the last full glottal pulse.
▪ A full glottal pulse is one that extends from the
baseline at least through F2 of the display.
▫ distinguishes it from the glottal pulses seen in
voiced stops, affricates, and fricatives, as well as some low-intensity pulses that extend only through f1

Question 19

Segmentation of Vowels & Nasals

Answer 19

When vowels are located before or after nasals, the change from an oral to nasal filter function (a murmur) can be used to find a vowel-nasal or nasal-vowel boundary
▫sudden change in intensity at the boundary between a nasal and a vowel and the sudden appearance of the low-frequency F1 characteristic of nasal cavity resonance

Question 20

Segmentation of Vowels & Semivowels

Answer 20

When vowels are located before or after semivowels (/ɹ,l, w,j/) or diphthongs (/ɑɪ, ↄɪ, ɑʊ, eɪ, oʊ/) the segmentation problem is very difficult because there are no natural boundaries.
▫Examples: “yellow” (/jεloʊ/) or “I honor” (/ɑɪjɑnɚ/)

Question 21

Segmentation of Obstruents

Often straightforward to segment obstruent

Answer 21

Often straightforward to segment obstruents
▪Stops: the offset (or end) of the closure interval is taken as the burst (same for the closure offset of affricates)
▪Fricatives: the offset is taken as the end of the frication noise, or the first full glottal pulse of the following vowel

Question 22

Segmentation of Obstruents

When two fricatives follow each other,

Answer 22

When two fricatives follow each other, the changing spectrum must be used to identify the end of one and beginning of the other
▫Can be challenging
▪When two stops follow each other, the only way to distinguish the closure intervals is if the first stop is released, producing a burst.

Question 23

Suprasegmentals

Answer 23

Prosodic variables
▪Fundamental Frequency (F0) contour
▫Across segments and sentence-level utterances
▪Can be computed by pitch tracker analysis programs
▪General declination F0 pattern for declarative utterances
▪Fall-rise pattern of F0 common for question utterances

Question 24

Suprasegmentals

Shape of an F0 contour carries

Answer 24

Shape of an F0 contour carries grammatical and affective information
▪F0 variation conveys emotional state by providing signals concerning the rules of turn-taking and by marking important aspects of an individual’s personality

Question 25

Tonal Languages

Answer 25

Examples: Thai, Igala, Mandarin, Cantonese
▪Shape of the F0 contour across a vocalic segment has phonemic contrastive function
▫in Mandarin the sequence /bɑ/ produced with a flat F0 contour means “eight,” but /bɑ/ spoken with a rapid fall-rise contour means “to hold

Question 26

Digital Analysis of Speech

Answer 26

Goal is to produce an accurate spectral analysis as a function of time. (similar to spectrogram)
▪There are many different algorithms and hardware solutions for this purpose
▪Original spectrographs used a continuous, analog process
▪Computer analysis of speech requires a conversion of analog speech waveforms to digital representation

Question 27

Speech Analysis by Computer: From Recording to Analysis to Output

Answer 27

All computers take an analog signal and digitize it, converting it to a series of discrete numbers in binary code (sequences of zeros and ones)
▪The analog signal is digitized with filtering at a specific sampling rate and bit level

Question 28

Sampling rate

Answer 28

Sampling rate: number of times per second the computer takes a “snapshot” of the signal and stores it in digital form
▫Standard sampling rate is 44.1 kHz
▫Highest signal frequency that can be analyzed reliably is one-half the sampling rate.

Question 29

Filters

Answer 29

Filters: eliminates all freqs above one-half of the sampling rate
▫Anti-aliasing filters

Question 30

Bit Rate

Answer 30

Bit Rate: number of amplitude levels available for storing amplitude variations in an analog signal
▫process of coding amplitude information during A to D conversion is called quantization

Question 31

Digital Parameters for Speech Analysis

Answer 31

Speech analysis programs have algorithms for display, editing, and analysis of speech waveforms.
▪Linear predictive coding: used to generate spectra to estimate formant frequencies
▪Programs can make errors during LPC analysis and F0 analysis, especially with clinical populations (e.g., hypernasality, breathy voice)

Question 32

Review

Speech waveforms provide information

Answer 32

Speech waveforms provide information on amplitude fluctuations as a function of time and fundamental frequency (F0) for voiced sounds, but do not provide direct access to formant frequencies
▪The spectrogram shows formant frequencies as a function of time & allows a user to infer changes in vocal tract configuration resulting from movement of the articulators.

Question 33

Review

It is possible to segment phonemic and phonetic

Answer 33

It is possible to segment phonemic and phonetic information from spectrograms
▫segmental durations, vowel formant frequencies, obstruent spectral characteristics,
▪Can analyze suprasegmental characteristics such as F0 and intensity contours
▪Computer-based speech analysis requires consideration of sampling rate, anti-aliasing filters, and quantization.

Question 34

Review

Analysis of formant frequencies is

Answer 34

Analysis of formant frequencies is generally fast, reliable, automatic, and accurate except in cases where a speaker has a breathy voice quality and/or excessive nasality.