In some cases the number of sentences that may be played back is too large to be recorded. For
example if the application deals with flight services then the system should be able to give
information about flight numbers such as: ``Flight A F Nine Three Zero One from PARIS will
land at ELEVEN HUNDRED TWENTY FIVE.''
It is obvious that the storage capacity and the
recording effort necessary to obtain all the possible combinations of
sentences are important. So the approach consists of recording
independent speech segments such as: ``Flight'', ``from'', ``will land
at'', names of all
the quoted cities, the digits, and the alphabet. The sentence to play back is obtained by linking appropriate
segments together through a substitution of individually stored words in the
information slots of the carrier sentence:
This is usual in applications like schedules, bank balances, etc.
The scenario to perform consists of listing all the sentences that may occur. The application developer should know whether the technology provider supplies any tool that generates all the possible sentences (generation of a written version of the sentences to check). He should be able to select the words to be individually stored (the variable parts) from the carrier sentences (the fixed ones). Afterwards one has to define the recombination rules to account for the language specific characteristics (assimilation of adjacent words, coarticulation effects, etc.). For example in some languages like French there is a particular liaison that imposes contextual rules and exceptions: the number ``21'' is pronounced /v~ te ~ / and cannot be produced by simply concatenating 20 and 1 (/v~ e ~ /). The corresponding rules should be delivered by the technology provider or implemented by the application developer using a suitable development environment.