Improving speech synthesis quality for voices created from an audiobook database

Abstract

This paper describes an approach to improving synthesized speech quality for voices created by using an audio book database. The data consist of a large amount of read speech by one speaker, which we matched with the corresponding book texts. The main problems with such a database are the following. First, the recordings were made at different times under different acoustic conditions, and the speaker reads the text with a variety of intonations and accents, which leads to very high voice parameter variability. Second, automatic techniques for sound file labeling make more errors due to the large variability of the database, especially as there can be mismatches between the text and the corresponding sound files. These problems dramatically affect speech synthesis quality, so a robust method for solving them is vital for voices created using audio books. The approach described in the paper is based on statistical models of voice parameters and special algorithms of speech element concatenation and modification. Listening tests show that it strongly improves synthesized speech quality.