Emilia: A speech corpus for Argentine Spanish text to speech synthesis

Abstract

This paper introduces Emilia, a speech corpus created to build a female voice in Spanish spoken in Buenos Aires for the Aromo text-to-speech system. Aromo is a unit selection text-to-speech system, which employs diphones as units of synthesis. The key requirements and design criteria for Emilia were: to synthesize any text in Spanish into high-quality speech with a minimum corpus size. The text corpus was designed to guarantee the phonetic and prosodic coverage. A three-stage strategy was used: in the first stage, 741 sentences were designed with all of the syllables of Spanish spoken in Argentina, with and without stress, and in all positions within the word; in the second stage, 852 sentences were added to balance out the distribution of the diphones; and after a perceptual evaluation of the quality of synthesized speech, in the third and final stage, 625 sentences were added to achieve the specified unit coverage, and to introduce sentences with more complex syntactic and prosodic structures. Issues from all three corpus building stages are reported. The paper also presents the results from the quality perceptual evaluations of the synthesized voice. Emilia has a duration of three hours and 15 minutes; its speech quality synthesized with Aromo system is similar to the level obtained with commercial systems, with a real-time ratio less than one.