Databases for Automatic Speech Recognition and Text to Speech Systems of
Argentine Spanish 

Gurlekian Jorge*, Colantoni Laura*, Torres Humberto*, Rodríguez Hernán*,
Rincón Antonio**, Moreno Asunción** and  Mariño José**.

*Laboratorio de Investigaciones Sensoriales CONICET. Buenos Aires. Argentina
** Applied Technologies on Language and Speech. Barcelona. Spain.


Two different databases of Argentine Spanish are presented here, one for an
automatic speech recognition (ASR) system, and the other for a text to
speech (TTS) system. 

The goal of the first project was the design and creation of a database to
be used in a ASR system under a fixed telephone network. One thousand
speakers, native to five Argentine dialectal regions were recorded. Each
speaker had to answer five questions and read 38 texts, which consisted of
numbers, last names and corporation names, as well as phonetically rich
sentences. Two hundred sets of nine sentences were selected from a pool of
7000 sentences, which were used to generate 1,000 prompt sheets. These
sentences contained all the contextual allophones transcribed with the
SAMPA alphabet.  A strategy of uniform information getting and recording
was designed. Phone calls were received through a digital line connected to
a computer, and monitored by an acquisition software. This procedure
resulted in an efficient way to collect data from different dialects in
only two months.

The second project involved the design of a relational SQL-based database
to generate an intonational model for an Argentine Spanish TTS system. The
first stage in the population of the database involved the massive load of
text, divided into three co-indexed files: sentences, orthographic words
and phonological syllables. A software tool, which performed phonemic
transcription and syllabic segmentation of the text, was developed to allow
indexation.  In the beginning, a large amount of sentences was loaded,
then, a minimum set of 741 sentences was selected, according to criteria
related to syllable occurrences in all positions in a word with and without
stress. This set contained 97% of all Spanish syllables extracted from a
widely used Spanish dictionary.  The utterances were recorded at 16KHz /
16bits, using an interactive program. Two professional announcers were
instructed to generate a variety of accent patterns and intonational
phrases to prevent monotony. Speech signals were then labeled by trained
phoneticians with a spectral analysis tool, using ToBI tiers (Beckman &
Ayers, 1994).  ToBI conventions were reviewed to account for the prosodic
patterns of Argentine Spanish. Frequency values were scaled using the ERB
scale, and bitonal accents were redefined.  Finally, the second stage of
the population of the database consisted of the incorporation of the
labeled files. Waveforms were kept outside the database but linked to it,
to allow the identification, reproduction, and selection of specific
segments.