۴۷۶۸پ
per
استخراج قسمتهای قابل استفاده در سیگنال گفتار هم-کانال برای بهبود عملکرد الگوریتم-های شناسایی گوینده
/اعظم خلیلی
تبریز، دانشگاه تبریز ،دانشکده مهندسی برق و کامپیوتر ، گروه مخابرات
۹۱ص
چاپی
فاقد اطلاعات کامل
کارشناسی ارشد
مهندسی برق-مخابرات - سیستم
۱۳۸۶/۰۶/۲۸
تبریز، دانشگاه تبریز ،دانشکده مهندسی برق و کامپیوتر ، گروه مخابرات
Speech processing-based algorithms are widely used in many applications such as speaker identification. Nevertheless, the performances of these algorithms are dramatically decreases in co-channel condition. Recently, to address this problem "usable speech" concept is introduced. Speech segments that are still usable for speech processing algorithms are known as "usable" speech. Since the usable speech is a context-dependent concept, so we consider usable speech extraction in speaker identification applications. In this thesis, we proposed a discrete wavelet transform (DWT) based algorithm to detect the periodicity properties of speech frames and used it in two cases. First, in voiced frames detection algorithm and to define a new criteria for usability of co-channel speech. Second, in usable speech extraction algorithm with a new threshold level. The results show that the proposed algorithm has ۵ improvement in performance compared with other algorithms. In addition, we used the measure to find the classifier that their fusion with the proposed algorithm for usable speech detection can lead to better performance. After finding the satiable classifiers we proposed a simple yet effective fusion algorithm to combine the decisions of classifiers. The simulation results for the proposed algorithms are evidence of suitability of theses algorithms
Co-channel speech
usable speech
discrete wavelet transform
speaker identification
confusion matrix
fusion algorithm
خلیلی، اعظم
طینتی، محمدعلی، استاد راهنما
مظفری تازه‌کند، بهزاد، استادمشاور
سیاه و سفید
نمایه‌سازی قبلی
