Why would we want to enroll specific users?
What is this called, and what does it represent?
How is the spectrum of the line spectrum analogous to the spectrum that comes out of a prism.?
What does Fourier's theorem say?
What is the wavy line described in the figure called?
What do the veritical lines signify?
What is a formant?
What kinds of voice sounds are formants good at characterizing?
Vowels are characterized as being front/back, and ....
How can formants be used to characterize vowel sounds?
What does each vertical line in the spectrogram indicate?
The darkness of each point on the spectrogram indicates the .... of a given sample at a given ....
By definition, stops are practically silent. What can we look at on the spectrogram that distinguishes one stop from another?
When we look a voice spectrogram, what is unique about the fricatives?
What is the input to a spectrogram?
Samples for a spectrogram are taken:
On the order of tenths of a second.
On the order of hundreds of a second.
On the order of thousandths of a second.
On the order of millionths of a second.
Each position in a vector has .....
True or false: distance between vectors can only be calculated for three dimensional space.
True or false: experienced mathematicians can close their eyes and visualize 5-dimensional space.
Each sample in the sectrogram is a line spectrum; a line spectrum is a vector where each dimension represents ....
Why do we want to gather up a number of subsequent samples in the spectrogram into larger segments when we're recognizing sounds?
Name three features that we can extract from a segment, and indicate what kinds of sounds they help to characterize.
Describe how a segment of sound input can be matched to a template.
If our input is a sequence of voice segments, the result of mapping each segment to its closest template is a sequence of .... called codes.
The illustration shows the codebook for our toy 'dad recognizer'. Describe how we would find out if a segment maps unto the symbol '/'.
Why is this called a finite state machine?
What's the main advantage of a finite-state technique to processing language?
The illustration shows a deterministic version of our toy 'dad recognizer'. Why is it called deterministic?
The illustration shows the Hidden Markov Model version of our 'dad recognizer'. Is this a deterministic model?
Is the illustration at right a finite-state model?
A probabilistic model (chose all that apply):
Where do we get the probabilities that go into a probabilistic model?
Why do we call it a model?
Say our acoustic HMM narrows one word of voice input down to a 50/50 choice between 'big' and 'pig'. What can we do to help us decide which it was?