Lecture 5 & 6

Question 1

What are 4 acoustic differences between the speech of males and females? What is the source of each difference?

Answer 1

Females have a higher F0 (shorter vocal cords)
Females have an increased vowel space (higher F0 = higher resonance bands)
Females have increased breathiness (vocal folds remain open for longer during glottal pulsing).
Females have a relatively more dominant fundamental frequency

Question 2

Why should a bandwidth span at least two harmonics during spectral analysis?

Answer 2

So that harmonics aren’t mistaken for formants.

Question 3

Why should a bandwidth be changed for males’ vs. females’ speech?

Answer 3

Because the value that is twice the harmonic value in males (e.g. 300Hz) may not be twice the harmonics in females (i.e. their F0 is above 150Hz). A bandwidth that is too narrow will resolve individual harmonics, and these may be confused with formants.

Question 4

Acoustically, what characterises infants’ vocalisations?

Answer 4

A very HIGH F0 and highly VARAIABLE F0

A very large bandwidth

Question 5

What kind of developmental changes affect the acoustic characteristics of children’s speech?

Answer 5

1. Biological/anatomical
2. Neurological maturation
3. Development/refinement of speech motor control and language processing (e.g. phonological knowledge)

Question 6

In what ways does the speech of children mature?

Answer 6

1. Fundamental frequency decreases
2. Average duration decreases (as they become faster/more efficient in production)
3. Variability in production decreases

Question 7

What are the main acoustic differences between speech spoken clearly (with care) and more informal, conversational speech?

Answer 7

1. Slower speaking rate, greater pausing, longer segment durations
2. Fewer reduced forms of consonants (i.e. final stops are more likely to be realised)
3. Intensity is greater for some times of consonants, and therefore is acoustically more distinctive.

Question 8

What is hyperspeech?

Answer 8

When the listening environment is demanding (i.e. high levels of background noise) then more resources are devoted to articulation to make speech more clear/distinctive.

Question 9

What is hypospeech?

Answer 9

When the listening demands are low (i.e. quiet environment) then fewer resources are applied to articulation. Hypospeech is characterised by reduced effort, and reduced deviation of the articulators from normal.

Question 10

How can we apply the hyper- hypospeech tradeoff to children with intelligibility issues?

Answer 10

Children with reduced resource capacity may be biasing hypospeech habitually because of concurrent resource demands in other cognitive domains (e.g., comprehension, semantics, working memory).

Question 11

How must we consider this resource capacity/demand in therapy?

Answer 11

Consider the speaking context (e.g. reduce background noise) so that the child may devote maximum resources to output.
Encourage child to consciously utilise the hyperspeech end of their articulatory continuum to a greater extent, e.g. minimal pairs (to highlight the pragmatic consequences of using misarticulated speech)

Question 12

What is the difference between intonation and prosody?

Answer 12

Intonation = modulation of Fø 
Prosody = a broader concept relating to intonation rhythm, rate and intensity variations.

Question 13

How is prosody measured acoustically?

Answer 13

F0 contour, amplitude contour, and duration

Question 14

How is stress realised acoustically?

Answer 14

1. Longer duration
2. Increase in 
fundamental frequency
3. Increase in amplitude.

Question 15

What are the two main steps in text-to-speech synthesis?

Answer 15

1. Text to phoneme conversion

2. Speech synthesis

Question 16

Describe text to phoneme conversion

Answer 16

Words and punctuation are converted to a string of phoneme symbols that are:

1. Prosodically marked to specify stress, intonation and duration variations).
2. Syntactically marked to disambiguate words like “object”.

Question 17

Describe phoneme to speech conversion

Answer 17

Phoneme symbols are computed to a digital sound wave, which is converted (via A-D conversion) to an electrical current that drives a loudspeaker.

Question 18

What are the two main types of speech synthesisers?

Answer 18

Formant synthesis and concatenated speech

Question 19

Describe formant synthesis

Answer 19

Two types of sound scources are used: voice pulses (e.g. for vowels) and noise (e.g. for fricatives)
Each phoneme is specified by its acoustic features (fundamental frequency, duration, noise amplitude etc.).
Certain filters are applied in order to produce the specified sound.

Question 20

How are rules specified in formant synthesis?

Answer 20

Rules are set automatically, and include contextual constraints (i.e. adjacent phonemes and sentence prosody). For example, formant frequencies of a vowel will change in a particular direction according to the following vowel.

Question 21

How often are the acoustic parameters specified in formant synthesis? Why?

Answer 21

Every 5-6ms , to reflect the dynamic changes in the spectrum that are characteristic of human speech

Question 22

What is an advantage of formant synthesis?

Answer 22

1. Relatively high efficiency (rules do not take up a large amount of memory)
2. Fast output rates of up to 600 words per minute
3. Highly flexible – can provide a reasonable pronunciation of unknown words
4. Reasonable intelligibility

Question 23

What is a disadvantage of formant synthesis?

Answer 23

1. Not natural sounding – sounds robotic (particular difficulty with prosody)
2. Development time is long (because of the large number of rules to be created) and therefore the products are expensive

Question 24

Describe concatenation speech synthesis.

Answer 24

This approach stores, concatenates (links together) and smooths sections of PRE-RECORDED speech. Digital recordings of an actual speaker are broken up into small units, which can be reconstituted into larger novel utterances.

Question 25

Is the phoneme a useful unit of speech to concatenate? Why?

Answer 25

No – because the number of possible transitions from one phoneme to another is large (and doesn’t result in a smooth transition)

Question 26

What units of speech are more successful to concatenate?

Answer 26

Syllables, diphones, and disyllables (as well as individual words in some systems). A fairly manageable number of digitally recorded syllables can be stored and combined into virtually any English utterance. Syllables are also useful because they have transitions built into them.

Question 27

What are diphones?

Answer 27

Diphones are formed by placing cuts in recorded CV or VC syllables at the steady state vowel component. All words that share a particular phoneme sequence use the same diphone for that sequence. Diphones are recombined to form longer utterances. E.g. the words ‘cat’ and ‘cab’ use the same diphone to produce /ca/. The next diphone would differ though: cat would use /at/ and cab would use /ab/. Therefore the production of the /a/ sound is based on the combination of the two diphones.

Question 28

How many diphones are needed to synthesise English speech?

Answer 28

About 2000

Question 29

What are disyllables?

Answer 29

They are half a syllable – either the onset up to the middle of the vowel, or from the middle of the vowel to the end of the coda.

Question 30

What is an advantage of using concatenation synthesised speech?

Answer 30

It is more natural sounding and intelligible than formant speech synthesis. They have shorter development times, and are relatively less expensive

Question 31

What is a disadvantage of using concatenation synthesised speech?

Answer 31

It requires a great deal of memory storage (although this is becoming less of a problem with the memory capacity of modern computers. Prosody is just as unnatural as formant synthesis, as well as discontinuities in the middle of vowels and between syllables.

Question 32

Does perceiving synthesised speech require more or less cognitive effort on behalf of the listener?

Answer 32

More

Question 33

What are some strategies to enhance the intelligibility of synthesised speech?

Answer 33

Slow down the speech rate, include visual texts

Question 34

In synthesised speech, what kinds of grammatical structures are perceived best?

Answer 34

Sentences are understood better than single words (indicating top-down processing). Simple sentences with high probability words are better understood than grammatically complex sentences.

Question 35

Give four examples of the clinical application of synthesised speech

Answer 35

1. Reading instruction – useful for children and adults with developmental or acquired reading disorders
2. Communication aids – people with speech production disabilities (e.g. dysarthria) can type a message, which will be spoken aloud
3. Multilingual communication systems – which require speech recognition, linguistic translation, and speech synthesis – still in early stages
4. Research – the researcher can control acoustic features of speech, and therefore investigate the acoustic cues used during speech perception

Question 36

What is a speech recognition system?

Answer 36

Convert spoken utterances into phonemic symbols and then to text in a word processing document.

Question 37

What is a speech understanding system?

Answer 37

Goes further than a speech recognition system by taking the words and analysing semantics

Question 38

What is a speech language system?

Answer 38

Speech language systems incorporate speech recognition, understanding and speech synthesis to provide a communicative interaction with a person.