Text-To-Speech Conversion

Ideally the input to a phonetic speech synthesizer should be a string of ASCII characters that is readable by a human. In order to provide complete text-to-speech capabilities, TTSCF uses a multi step process that converts ASCII text into natural sounding speech waveforms. Individual steps of this process are:

1. "Normalization:" The text normalizer takes the ASCII input and expands abbreviations, numbers and monetary amounts to their full word form. It also processes punctuation, nonalphabetic input characters and special pronunciation.
2. "Syntactic Analysis:" A crude syntactic analyses of the sentence is performed based on the syntactic role of function words and verbs that were detected.
3. "Phonemic Translation:" Incoming text is compared against an exception dictionary. If a match for a word is found, the associated phoneme string and stress information are retrieved and passed directly to the next process. If the dictionary search fails to find a match, the word is assigned a phoneme string and stress pattern based on an extensive set of rules that are similar to the rules used for reading English aloud. The mechanism of assigning phonemic and stress information to a word is called synthesis-by-rule.
4. "Parameter Generation:" Given the phoneme string, lexical stress and syntactic information, appropriate parameters should be generated and supplied to the acoustic synthesizer.

Conversion Layers

The following table provides a list of the functions used for conversion required for transformation of text to speech.

Table 7.2: Text-To-Speech Conversion Facilities

Input	Function	Output
eng_string	x_eng2ipa	ipa_string
ipa_string	x_ipa2sca	sca_string
sca	sca2b	scb
scb	scb2a	sca
sca_string	sca_say	VOICE
scb_string	scb_say	VOICE
sca	sca_utter	VOICE
scb	scb_utter	VOICE

Exception Dictionaries

Two user maintained phonetic dictionaries are used by TTSCF software. These phonetic dictionaries may be used to conversion rules when desired.

English to IPA Dictionary

English to IPA dictionary is a relation that contains tuples of the form: <"english-word": "ipa-string">. This dictionary can be maintained by the "Binary Table Editor".

English to SCA Dictionary

English to SCA dictionary is a relation that contains tuples of the form: <"english-word": "sca-string">. This dictionary can be maintained by the "Binary Table Editor".

English to IPA Conversion Rules

English to IPA conversion is done through a set of phonetic rules published by the Naval Research Laboratories. Each rule has the form:

    [left] [match] [right]   [out]