Speech Recognition

Question 1

Segmentation is one problem listeners with a speech signal have. What is it?

Answer 1

Dividing the speech input into phonemes (units of sounds) and words

Question 2

What is coarticulation?

Answer 2

Speech signal is variable because pronunciation of a phoneme depends on speaker’s pronunciation of preceding and following phonemes

Question 3

What are speech signals due to?

Answer 3

Differences in speaker’s sex, dialect and speaking rate

Question 4

How phonemes do speakers typically produce per second?

Answer 4

Around 10 - listeners must identify what is being said very rapidly. Non-native speakers often produce many speech errors

Question 5

What is energetic masking?

Answer 5

Speech signal is hard to perceive due to distracting sounds e.g. other speakers

Question 6

What are the 3 cue categories mattys et al identified in the hierarchical approach to segmentation?

Answer 6

1. Lexical, syntax, word knowledge -> used if listening conditions are good
2. Segmental -> coarticulation
3. Metrical prosody, word stress -> used only if other cues cannot be used

Question 7

What is speaker’s variability?

Answer 7

Listeners use speaker characteristics e.g. American accent to form a speaker model. This speaker model then influence how listeners interpret speech signal

Question 8

add in info from graph