Applied Phonetics Portuguese Text to Speech Arlo Faria University of California Berkeley Linguistics 110 Prof Ian Maddieson Contents 1 Introduction 1 2 The phonetics of Portuguese 2 1 Historical Background and Overview 2 2 Vowels 2 2 1 The principal vowels 2 2 2 Diphthongs and Triphthongs 2 2 3 Nasal vowels 2 3 Consonants 2 4 Syllable Structure 2 4 1 Onsets 2 4 2 Rhymes 2 5 Lexical Stress 3 The orthography of Portuguese 3 1 Vowels 3 1 1 Distribution of high mid and low mid 3 1 2 A note on vowels and accent marks 3 1 3 Diphthongs 3 2 Consonants 3 3 Nasalization 3 4 Syllable structure 3 5 Lexical Stress 2 3 3 3 5 6 6 7 7 8 8 vowels 10 10 11 13 13 14 16 17 18 20 20 21 22 22 23 24 4 Text to speech Implementation 4 1 Components of text to speech synthesis 4 2 Text preprocessing 4 3 Word pronunciation 4 3 1 Letter to phone mapping 4 3 2 Syllable parsing and stress assignment 4 4 Prosody and Signal Processing 5 Conclusion 24 A Recorded Data A 1 Recording Process A 2 Measurement of vowels A 3 Diphthongs and Nasals A 4 Consonants A 5 Stress A 6 High mid vs low mid vowel perception 28 28 29 30 30 30 32 Applied Phonetics Portuguese Text to Speech Arlo Faria University of California Berkeley Linguistics 110 Prof Ian Maddieson May 16 2003 Abstract This paper describes a text to speech application for a variety of Brazilian Portuguese After presenting the language s phonetic attributes the orthographic system is examined and shown to be a function that maps letters to these sounds Given the orthography s phonological regularity it is simple to implement the textual analysis portion of a speech synthesis system as I demonstrate with some simple Perl code 1 Introduction For decades technologists have been predicting that computers will radically revolutionize the way we live Visionaries have long dreamed of a future in which machines are integrated into nearly every facet of daily life From artificially intelligent robotic agents to embedded microcomputers in household appliances it will be virtually impossible to do anything that is not somehow wired As the power of computing grows exponentially no one is even able to imagine the possible applications At least one thing is certain human computer interaction will be far more intricate than today s mouse keyboard and monitor interface More likely users will want to relate to machines in a more natural manner in a more human manner In this respect computers of the future will need to be capable of receiving and conveying information through our most preferred channel of communication speech Robust speech recognition and naturalsounding speech synthesis will be part of the new interface protocol Redefining input and output is just one of the many promising applications of speech and language processing that have only recently become possible Other examples of speech technologies learning foreign languages1 automated telephonic customer service2 and assistive technologies for the disabled3 This paper presents the framework for a text to speech system for a common variety of 1 Computer Aided Language Learning http www ocf berkeley edu arlo CALL pdf http www tellme com 3 http www perl com pub a 2001 08 27 bjornstad html 2 1 2 2 THE PHONETICS OF PORTUGUESE Brazilian Portuguese Sparing explicit technical specification it highlights the linguistic background that is necessary for such a task Taking advantage of the language s phonological orthography the process of translating from letter tokens into speech segments is explored in great detail I first discuss the phonetic attributes of Portuguese describing the inventory of sounds along with the language s syllable structure and stress pattern To illustrate these features a set of acoustic and auditory measurements are provided Some of these word lists and figures are included in the Appendix The recorded speech was preserved as digitized sound files available online www ocf berkeley edu arlo ling110 Instead of a treatment of Portuguese phonology I then proceed to explain the Portuguese orthographic system Indeed the orthography is very phonological the information needed to derive a phonetic representation is almost entirely contained in the orthography s set of letter to phone rules These rules are listed thoroughly defining the function mapping a sequence of letters into a sequence of sound segments Portuguese syllable structure and lexical stress are examined and it is shown that these are also represented in the orthography The final portion of this paper is a demonstration of how to apply the phonetic background and orthography of Portuguese in the implementation of a speech synthesis system Specifically the phonological regularity of the Portuguese orthography greatly facilitates the primary stages of the text to speech process An illustrative example is provided with the letter to phone rules being straightforwardly coded into Perl regular expressions The text to speech implementation that I describe proceeds satisfactorily until the intermediate stage of the process it is in effect a system that translates text into its phonetic realization For practicality I show an application that processes a user s text input and displays its reading in IPA transcription Such a text to transcription tool has considerable accuracy not common for many contemporary speech technologies and could be very useful to linguists and learners of Portuguese Finally I discuss how one could complete the text to speech application by constructing an appropriate audio signal from the phonetic representation It is evident that this final step presents the greatest technical challenge and I mention a simple approach to speech synthesis This paper thoroughly details the first two of four steps in a text to speech system Building upon this work the last two prosodic analysis and signal processing are areas which I hope to one day explore 2 The phonetics of Portuguese Underlying any implementation of speech technology is a thorough understanding of the linguistic systems of the language being targeted For this Portuguese text to speech system it is necessary to be acquainted with the sounds and patterns of a particular variety of Brazilian Portuguese This section describes the phonetic inventory utilized in the speech of the author Arlo Faria 2 1 Historical Background and Overview 2 1 3 Historical Background and Overview Portuguese is a Romance language related to Castilian and Catalan A unique type of Latin that was spoken in