Phonologic-graphemic transcodifier for Portuguese Language spoken in Brazil (PLB)
Abstract
An automatic speech-to-text transformer system, suited to unlimited vocabulary, is presented. The basic acoustic unit considered are the allophones of the phonemes corresponding to the Portuguese language spoken in Brazil (PLB). The input to the system is a phonetic sequence, from a former step of isolated word recognition of slowly spoken speech. In a first stage, the system eliminates phonetic elements that don't belong to PLB. Using knowledge sources such as phonetics, phonology, orthography, and PLB specific lexicon, the output is a sequence of written words, ordered by probabilistic criterion that constitutes the set of graphemic possibilities to that input sequence. Pronunciation differences of some regions of Brazil are considered, but only those that cause differences in phonological transcription, because those of phonetic level are absorbed, during the transformation to phonological level. In the final stage, all possible written words are analyzed for orthography and grammar point of view, to eliminate the incorrect ones.
- Publication:
-
NASA STI/Recon Technical Report N
- Pub Date:
- 1994
- Bibcode:
- 1994STIN...9517649F
- Keywords:
-
- Languages;
- Orthography;
- Phonemes;
- Phonetics;
- Speech Recognition;
- Verbal Communication;
- Brazil;
- Grammars;
- Phonemics;
- Words (Language);
- Communications and Radar