خط مشی دسترسیدرباره ماپشتیبانی آنلاین
ثبت نامثبت نام
راهنماراهنما
فارسی
ورودورود
صفحه اصلیصفحه اصلی
جستجوی مدارک
تمام متن
منابع دیجیتالی
رکورد قبلیرکورد بعدی
Document Type:Latin Dissertation
Language of Document:English
Record Number:52917
Doc. No:TL22871
Call number:‭1484454‬
Main Entry:Mahnoosh Mehrabani
Title & Author:Automatic dialect separation assessmentMahnoosh Mehrabani
College:The University of Texas at Dallas
Date:2009
Degree:M.S.
student score:2009
Page No:54
Abstract:Dialect variations of a language represent considerable challenges for sustained performance of speech systems. In a given language space, estimation of similarity or diversity between multiple dialects provides valuable information for speech researchers. Assessing dialect proximity or separation knowledge can be used for improving automatic speech recognition as well as speech coding or speaker recognition systems, allowing conservation of resources or predicting performance if new dialects are introduced. Knowing the separation between dialects of a language can also be useful for assessment of training data for a dialect classification system. Despite the benefits of dialect proximity assessment and research interest in this field, virtually no studies have addressed this issue. In the present study, fundamental differences between dialects are explored. First, a method is proposed to measure spectral acoustic differences between dialects based on comparing log-likelihood score distributions, using traditional MFCC features and GMM models. Next, text-independent prosody features based on pitch and energy contour primitives are proposed to study the excitation structure differences between dialects. The proposed dialect proximity measures are evaluated and compared on a corpus of Arabic dialects and a corpus of South Indian languages which are closely related languages from the same language family, and are representative of similarities to pure dialect separation assessment. The proposed assessment framework employing measures of spectral, pitch, and energy structure are also shown to be consistent and repeatable. Finally, a formal subjective 3-way dialect assessment is performed to illustrate the relation between automatic system results and human perception.
Subject:Applied sciences; Electrical engineering; Artificial intelligence; 0544:Electrical engineering; 0800:Artificial intelligence
Added Entry:J. H. L. Hansen
Added Entry:The University of Texas at Dallas