SLPA 456 Exam 4

An analog signal is _____ and __ _______

An analog signal is _____ and __ _______

Continuous and time-varying

Speech is an example of a _________ signal

Analog

A digital signal is ______.

Discrete

3 main parameters of sound

frequency, time, and amplitude

3 types of errors that can occur during ADC

Jitter, Quantization noise, and Aliasing

Jitter:

deviation in periodicity

• can be a result of irregularities in sampling rate

Quantization noise:

deviation in amplitude measures

• can be result of rounding errors in process of quanization

Aliasing:

distortion due to misidentification of frequency

• can be result of inappropriate sampling rate

Digital Signal Processing (DSP)

Pre-Processing of a digital signal

Steps of DSP

Speech Signal

1. Filtering

2. Digitization

3. Frame Selection

4. Windowing

5. Short-term analysis

   1. Graphic display or numeric output

Elements of filtering

Pre-emphasis, presampling

Elements of digitization

time sampling, quantization

elements of Frame Selection

Frame length, frame overlap

elements of windowing

tapering function

elements of short-term analysis

FFT, LPC, Cepstrum

elements of Graphic display or numeric output

spectogram, spectrum, other

Goal of filtering:

retain wanted parts of the signal while removing parts that do not necessarily provide any information

Pre-Sampling:

“anti-aliasing” - applying filters that block frequencies above the Nyquist frequency for that sample

Aliasing

underrepresentation of the sampling rate because the original signal is underrepresented

Example of anti-alias filter:

DC Off-set

Pre-Emphasis:

Equalizes (boosts weaker) energies over a specified range of frequencies so important aspects of signal have sufficient energy to accurately capture within the quantization bits available

Practical example of filtering in Aud and SLP

Measuring Auditory Brainstem Responses (ABR)

Removes:

• direct current (DC) signals from other electronic equipment

• 60 Hz hum from alternating current (AC) power sources

• background EEG activity, unwanted brain activity

  • uses pre-emphasis method called differential amplification

    • boosts level of desired evoked potential response while removing the extra noise.

Frame Selction/Windowing:

process of selecting which parts of signal to be analyzed

Window/Frame:

the portion of the signal selected to perform an analysis on

Windowing option examples:

Rectangle

Bartlett

Hanning

Hamming*

Blackman

Gauss

How is ABR recording windowed?

Based on a TIME-specific analysis!

Types of graphic displays of acoustic data:

Waveform, Spectrum, Spectogram, Profiles or contours

Dimensions of a waveform:

Amplitude by time

Types of waveforms (temporal analysis)

raw, envelope

Dimensions of a spectrum:

Amplitude by frequency

Types of spectrums (Spectral Analysis)

Fast Fourier Transform, Linear Prediction Coding, Cepstrum

Dimensions of spectogram:

Amplitude by frequency by time

Types of spectrograms (speech (complex) analysis

Conventional, countour, waterfall

Dimensions of Profiles or contours

Parameter by time

Types of profiles/contours:

f0 trace (pitch contour), intensity profile

Temporal (time-based) analysis works directly on the ______.

Waveform

What information can you analyze from a waveform?

Fundamental frequency

Perturbation Measures

Signal-to-noise ratio

Voice onset time

Vowel duration

Envelope

Fundamental Frequency:

frequency at which a system oscillates/resonates freely

Signal Processing Strategy used to get fundamental frequency:

Pitch determination algorithm (PDA) or pitch extractor

Temporal methods used by PDA:

Zero crossing

Peak Picking

Auto correlation (most modern)

Zero Crossing:

counts every time a wave passes through the zero line within a second, then divides by two to obtain the fundamental frequency

Peak Picking:

Fundamental frequency is derived by identifying wave peaks and counting either the total number of crests or troughs OR total number of peaks in general and dividing by 2

Perturbation measures:

3 types we can measure

• jitter

• shimmer

  • signal to noise ratio

Perturbgation:

a deviation from truly periodic and regular patterns of vibration of the vocal folds

Jitter:

variability in the fundamental period of phonation

• reported in an absolute value (ms) or relative value (%)

Jitter Percent:

obtained by dividing absolute jitter value by mean fundamental frequency period

Shimmer:

variability of amplitude of successive cycles of waveform

• reported in an absolute value (dB) or relative value (shimmer %)

Shimmer Percent:

obtained by dividing absolute shimmer value by the mean amplitude of the waveform

Signal to Noise Ratio:

Ratio of Periodic energy to aperiodic energy in the voice waveform

With NO background noise, SNR = _________

The intensity of the signal

When background noise is louder than the signal, SNR = ________

A negative value

Voice Onset Time:

duration of the interval between release of a stop consonant and the onset of vocal fold vibration (vowel production)

Vowel Duration:

duration of the interval over which the formant pattern (specifically F1 and F2) is stable

• aka vowel steady rate

Envelope:

overall profile of waveform

Spectral (frequency based) analysis operate directly on a _______

spectrum

Commonly used software for spectral analyses:

Audacity

PRAAT

Computerized Speech Lab (CSL)

Which spectral analysis software has few spectral analyses options?

Audacity

Which spectral analysis software is most widely used acoustic freeware?

PRAAT

Which spectral analysis software is professional software?

Computerized Speech Lab

Major types of Spectral Analysis:

Fourier Transform: Discrete (DFT) and Fast (FFT),

Linear Predictive Coding (LPC),

Cepstral based analyses,

Mel Frequency Cepstral Coefficients (MFCC)

Fourier Transform

Decomposes a waveform to reveal its frequency content to convert a waveform to a power spectrum

Discrete Fourier Transform

Fourier transform of a finite set of discrete samples from the waveform (determined by sampling rate and windowing)

• transforms data from samples into distinct frequency lines within a power spectrum

Fast Fourier Transform

optimized algorithm to calculate DFT

• all speech analyses software packages have an implementation of FFT

Linear Predictive Coding

Based on Quazi-periodic nature of speech, by knowing certain parts of the speech signal, other parts can be predicted

Cepstrum

A fourier transfer performed on the spectrum

• inverse/transposition of spectrum

What is a cepstrum useful in investigating?

Periodicity/ rate of change of a signal

Terms associated with Spectrum vs. Cepstrum:

Spectrum: frequency and amplitude → Harmonics → filtering

Cepstrum: Quefrency and amplitude → Rahmonics → liftering

2 important features of a cepstrum:

• preserves magnitude information about the signal and discard phase related info

  • emphasizes periodic nature of harmonics

What do cepstrum algorithms reveal in a signal?

Converting the signal and finding one formant enables algorithms that help find patterns to find the others

What do rahmonics show?

correlates to the perceptual “quality” measures of voice

Mel Frequency Cepstral Coefficients (MFCC)

represent short-term power within a second

• represents frequency bands as evenly spaced whereas cepstrum represents frequency bands linearly

  • more representative of human auditory sensitivity (perception of pitch)

Practically, when is mel frequency cepstral coefficients most useful?

in audio compression and speech recognition systems (eg. HA mapping)

How to obtain formants:

by using any spectral analysis method

Two main characteristics of formants:

• peak in spectrum of a vowel sound or energy bands in spectrogram

  • resonance of vocal tract

Which formants are typically used to describe most speech sounds?

F1 and F2

For vowels, what does F1 describe?

Tongue Height

For vowels, what does F2 describe?

tongue position

Formant Amplitude:

Relative amplitude of formants in a formant pattern?

Formant Space:

aka acoustic working space, acoustic vowel space, vowel triangle

• plot of F1 vs F2

• measures speech intelligibility

  • several other measures are derived from formant space.

Examples of measures based on (static) formant space:

• vowel space area

• formant centralization ratio

• four vowel articulation index

• Formant centroid

• Vocalic anatomical functional ratio

  • long-term formant distribution

Measurements based on “dynamic” aspect of formants:

• Formant Transition

• Formant Locus

• Formant Slope

  • Locus equation

Vowels, glides, and consonants differ in degree of ________.

Constriction

Sonorant Consonants

NO pressure build up at constriction

Nasal Consonants

lower the velum allowing airflow in nasal cavity

Continuant Consonants

do not block airflow in oral cavity

Resonators:

specific state of vocal tract that amplifies frequencies near the natural frequency of that system

Natural Frequency of a resonator is based on _____.

Length and diameter of the vocal tract

Relation of harmonic frequencies to resonating frequency

If close to resonating frequency: will be amplified

If far from resonating frequency: will be dampened

Relationship of two formants when they are close in frequency to one another,

They tend to boost each other’s amplitude

Formant Bandwidth:

difference (in Hz) between frequencies at +/- 3 dB of the intensity of the center frequency within a formant

Which graphic representation can you find formant bandwidth?

on a Spectrum

Practical use of formant space measurements:

represents maximum working space of a talker

• representative of maximum performance

Vowel Space Area

aka F1-F2 area

• calculated using a specific formula identifying the area of formant space graph

  • Used to study variety of speech and voice disorders

Long term formant distribution (LTF)

• average formant frequency of a given speaker

• calculated by taking average of all formants across all vowels in recorded sample

  • used to study variety of speech and voice disorders

Primary use of LTF:

forensic speaker identification and in studying effects of age and sex on speech

When is speech dynamic?

when there are changes as a result of consonants embedded along with vowels -- typical running speech

Formant transition:

relative shange from a vowel to a consonant

What speech sounds are formant transitions specifically associated with?

stop consonants

Formant locus:

characteristic value for each place of consonant articulation

** helpful to judge phonemes and speech intelligibility

Formant slope:

the change in formant frequency over an interval of formant transition

** helpful in studying speech intelligibility in dysarthric speakers