Next: Speaking Style
In some cases - very often in the scientific context or in
combination with speech synthesis - the contents of a speech corpus
have to be specified not in term of vocabulary but in terms of
phonological units, like phonemes, syllables, morphemes.
For instance, a general purpose speech recognition system will
require a minimum of repetitions of every possible phoneme in
various contexts by each speaker.
Or a corpus for concatenative speech synthesis will require every
diphone combination uttered from the same speaker in a minimum of 20
different left and right contexts.