Datasheet

3 - 4
A WORD ABOUT TEXT TO SPEECH
On the whole, the operation of the text-to-speech feature is very easy and
straightforward. But does this mean that the technology is as well? In fact, the
opposite is true! Let’s have a closer look at this exciting technology.
Converting text to intelligible sound is a complex process incorporating three
steps: a linguistic, a phonetic and an acoustic phase. Each module plays a distinct
roll in the text-to-speech process.
From Writing to Sounds
In the first stage, the linguistic module analyzes the recognized text.
Text normalisation is the first step. Punctuation symbols - comma’s, dots,
exclamation and question marks, hyphens etc. - are identified and abbreviations,
dates, telephone numbers, account numbers etc. are converted into plain text.
Depending on the context, the punctuation symbols can get pronounced: "/"
will be pronounced as "slash", ":" as "colon" etc. This only occurs when these
symbols have to be pronounced to get noticed. Pronouncing special symbols is
necessary to read structured character strings as "1245AS//876" but not in body
text such as "Shakespeare''s Globe" and "John said: ': 'I wont.'.'" A typical example
is reading telephone numbers such as "(800) USA--1234" where the "-" symbol is
pronounced as "dash".
It is interesting to note that in some languages - for instance English, French
and Dutch - the pronunciation of figures varies when you are dealing with a year
and not a “normal” figure! The number 1,,999 becomes "one thousand nine hun-
dred and ninety", the year 1999 is pronounced as "nineteen ninety six".
Next, all idiosyncrasies of the recognized text are identified to determine the
correct pronunciation. This means that the word "Dr." will be correctly inter-
preted as "doctor" and "drive" in sentences as “Dr. Spencer visited a patient on
Roosevelt Dr.".
All common abbreviations are solved with a specialized dictionary: the sys-
tem knows that "Corp." stands for "Corporation", "Bros." for "Brothers" etc. Lin-
4chapter3.p65 2/27/2001, 8:47 AM4