Homework on Ladefoged (2005) Chapter 9 ("Listening Computers")
Copyright © 2007 Grant McGuire and Mary E. Beckman
The assigned reading
The questions in this homework are in conjunction with
the assignment to read Chapter 9 of Ladefoged (2005)
on "Listening Computers".
Schedule
Your answers are due in class on
Monday, November 19, for discussion in class that day.
Instructions
Do the assigned reading described in the first section of
this document and then type your answers on a single sheet of
paper to bring to class with you to turn in. You may want to
make a second copy of this sheet of paper, to take notes as
we go over the answers.
Questions to answer
- Probability of a spectral pattern given a segment --
In Chapter 9, Peter Ladefoged talks about the role of probability
in different stages of speech recognition for different kinds of
tasks. He describes, for example, the way that an ASR system
might recognise the day of the week being pronounced
in the utterance in Figure 9.2.
Given the energy pattern, there is a high probability that
the patterns of slices over the first 150 milliseconds
of the signal will be better matches for the stored patterns
for an s or for an ʃ than for a w.
What stage of the speech recognition system in Figure 9.4
is involved when this pattern matching is done?
- Word probabilities (1) --
Describe how this high probability of an s or ʃ
match combines with the knowledge that the word is the name of a day
to further limit the choices.
What stage of the speech recognition system in Figure 9.4
would be involved if we knew for sure that the word
must be the name of a day of the week?
- Word probabilities (2) --
In making up the two pseudo-sentences
(1) Note that is the speaker's lips together with two
groups of a vowel in the air becomes too difficult when
lamenting.
and
(2) Same the without vocal but air as itself repetition
higher distinctiveness meaning major at standard absolute
can the consider.,
Peter Ladefoged relied on different words having different
probabilities. In one case he used raw probabilities and
in the other he used conditional probabilities. Which is which?
- Consonant and vowel probabilities --
What do the two barcharts in Figure 9.5 show?
What is the difference between
the barchart on the left and the barchart on the right in this
figure? Why are these numbers important in speech recognition?
Following up:
Ladefoged talks about the small number of occurrences of
ʒ and ʧ in the barchart on the
right, but is this surprising, given the small number of
occurrences of these two consonants in the barchart on the left?
Are there other sounds that have more suprising numbers on the right
given their numbers on the left?
Acknowledgments and references
The reading that is assigned for this homework is from the book:
Peter Ladefoged (2005). Vowels and Consonants: An
Introduction to the Sounds of Languages. 2nd Ed.
Blackwell.