Rule-Based Synthesis/Recognition Function

Rules for Phonemes + Rules for Variation

Machine Learning: Generative AI Synthesis Function

Finds patterns in the data to learn millions of rules

Rule-Based Synthesis/Recognition Pros & Cons

Small and Flexible, Science doesn’t know all of the rules, Synthesis sounds just okay, Recognition doesn’t work...

Machine Learning: Generative AI Synthesis Pros & Cons

Finds patterns in the data to learn millions of rules, Needs HUGE amounts of data, Sounds great

Complete Data for Generative Synthesis

Every phonemic environment, every suprasegmental environment, every dialect

Barriers to Generative Synthesis

Money, environmental resources, inflexible, majority favoritism, access to data collection, breadth of data, uneven data, profitable market

The Problem of Invariance

Speech tech can’t use the same acoustics for every phoneme/word but must take into account interspeaker and intraspeaker variation.

Intraspeaker Variation

the inability of a single speaker to produce a phoneme with the same acoustics twice

Suprasegmental Features

Answer Focus, Question vs. Declarative Intonation, Topic Marking, Emotions

a noncontrastive variant that only appears in certain environments

formants, harmonics, voiceless turbulent air

Rules For Formant Based Synthesis

individual phoneme goals, transitions between phonemes

m = [1] for p in [1, 2, 3]: m.append(p + m[p-1] + 2) print (m)

x = [400] for n in range(1,11): x.append(x[n-1] + (-.25*x[n-1] + 175)) for n in range(11,21): x.append(x[n-1] + (-.25*x[n-1] + 125)) print(x)

Phonemic variation in different environments

Vocal tract has intermediate shapes when moving between phonemes

Characteristics of Coarticulation

Formant transitions, the gestures of one phoneme overlapping with its neighbor, shape of phoneme adjusting for neighbors

Anything that moves during speech

Movement Change Approaching Goal

Movement slows approaching goal

Syllable length variation

All syllables are the same length

k in Dynamic Systems Equation

stiffness, acceleration of system

Current position equation

current position = previous position + (k*previous position + c)

Words that differ by only one phoneme, determine phonemic inventories

The study of which allophone a phoneme manifests as

A phoneme takes on the feature of a neighboring phoneme

Help

Options

focusNode

Didn't know it?
click below

Knew it?
click below

Don't Know

Remaining cards (0)

Know

retry

shuffle

restart

0:00

Embed Code - If you would like this activity on your web page, copy the script below and paste it into your web page.

Normal Size Small Size show me how

Ling Midterm 2

Term	Definition
Rule-Based Synthesis/Recognition Function	Rules for Phonemes + Rules for Variation
Machine Learning: Generative AI Synthesis Function	Finds patterns in the data to learn millions of rules
Rule-Based Synthesis/Recognition Pros & Cons	Small and Flexible, Science doesn’t know all of the rules, Synthesis sounds just okay, Recognition doesn’t work...
Machine Learning: Generative AI Synthesis Pros & Cons	Finds patterns in the data to learn millions of rules, Needs HUGE amounts of data, Sounds great
Complete Data for Generative Synthesis	Every phonemic environment, every suprasegmental environment, every dialect
Barriers to Generative Synthesis	Money, environmental resources, inflexible, majority favoritism, access to data collection, breadth of data, uneven data, profitable market
The Problem of Invariance	Speech tech can’t use the same acoustics for every phoneme/word but must take into account interspeaker and intraspeaker variation.
Intraspeaker Variation	the inability of a single speaker to produce a phoneme with the same acoustics twice
Suprasegmental Features	Answer Focus, Question vs. Declarative Intonation, Topic Marking, Emotions
Allophones	a noncontrastive variant that only appears in certain environments
Pure Tones Function	formants, harmonics, voiceless turbulent air
Rules For Formant Based Synthesis	individual phoneme goals, transitions between phonemes
Dynamical Systems Equation	Δx/Δt = kx + c
basic for loop	m = [1] for p in [1, 2, 3]: m.append(p + m[p-1] + 2) print (m)
Synthesis For Loop	x = [400] for n in range(1,11): x.append(x[n-1] + (-.25x[n-1] + 175)) for n in range(11,21): x.append(x[n-1] + (-.25x[n-1] + 125)) print(x)
Coarticulation	Phonemic variation in different environments
Formant Transitions	Vocal tract has intermediate shapes when moving between phonemes
Characteristics of Coarticulation	Formant transitions, the gestures of one phoneme overlapping with its neighbor, shape of phoneme adjusting for neighbors
Articulator	Anything that moves during speech
Movement Change Approaching Goal	Movement slows approaching goal
Syllable length variation	All syllables are the same length
Articulation System Type	Spring System
k in Dynamic Systems Equation	stiffness, acceleration of system
c in Dynamic Systems Equation	helps determine goal
x in Dynamic Systems Equation	given position of system
Current position equation	current position = previous position + (k*previous position + c)
Minimal Pairs	Words that differ by only one phoneme, determine phonemic inventories
Dialect	different phonemic inventories
Phonology	The study of which allophone a phoneme manifests as
Assimilation Rule	A phoneme takes on the feature of a neighboring phoneme

Created by: user-1886742

"Know" box contains:
Time elapsed:
Retries: