Phonetically Transcribed Speech Corpus Designed for Context Based European Portuguese TTS

Abstract. This paper presents a speech corpus for European Portuguese (EP), designed for context based text-to-speech (TTS) synthesis systems. The speech corpus is intended for small footprint engines and is composed by one sentence dedicated to each sequence of two phonemes of the language, incorporating as many language contexts as possible at diphone and word levels. The speech corpus is presented in three forms: its orthographic sentences, its phonetic tran-scriptions considering the words coarticulation, typical phenomenon in EP and its phonetic transcriptions considering the words coarticulation and the vocalic reduction effects, which is another typical phenomenon of EP language. The paper describes the corpus and presents the results in total number of phonemes and their distribution in the word and the sentence, and the total number of di-phones, considering and not considering the vocalic reduction for the phonetic transcription