Home // SOTICS 2014, The Fourth International Conference on Social Eco-Informatics // View article

An Audiobooks-based Approach for Creating a Speech Corpus for Acoustic Models

Authors:
Salvatore Michele Biondi
Vincenzo Catania
Raffaele Di Natale
Antonio Rosario Intilisano
Ylenia Cilano
Giuseppe Monteleone
Daniela Panno

Keywords: Automatic Speech Recognition; Audio Databases; Audiobook; Speech processing

Abstract:
Abstract—The limited availability of a valid speech corpus is one of the major problems affecting the design of speech recognition acoustic models. As a matter of fact, large amounts of manually-transcribed data is necessary in order to build a valid acoustic model. Nevertheless, obtaining large datasets is generally both time- and resource- consuming as it requires a continuous supervision of the entire building process. Speech corpora can be used to generate an acoustic model, however a large part of these are not suitable or freely available. This paper aims at showing the use of audiobooks as databases for creating speech corpora. An automatic algorithm that processes audiobooks for building speech corpora is proposed. This method allows to replace traditional manual transcription of audio- recordings and to automatically obtain a phonetic dictionary. An Italian acoustic and linguistic model was generated as use-case to test the effectiveness of the proposed procedure.

Pages: 1 to 5

Copyright: Copyright (c) IARIA, 2014

Publication date: October 12, 2014

Published in: conference

ISSN: 2326-9294

ISBN: 978-1-61208-372-8

Location: Nice, France

Dates: from October 12, 2014 to October 16, 2014