Speech modeling based on balanced reduction techniques

• المؤلف الرئيسي: Gao, Guangmin
• مؤلفون آخرون: Li, Gang
• التنسيق: Theses and Dissertations
• اللغة: English
• منشور في: 2008
• الموضوعات: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• الوصول للمادة أونلاين: http://hdl.handle.net/10356/13366

In this thesis an CELP algorithm based speech signal processing sys-tem is developed, in which, unlike the classical modeling where an Auto Regressive model is used, a stable Auto Regressive Moving Av-erage (ARM A) model employed. The basic idea is to first estimate the vocal tract filter with an Auto Regressive(AR) model of very high order and then convert it into an ARMA model via the powerful Balanced Model Reduction(BMR) techniques. Thus, the difficulties in the direct estimation of ARMA parameters is avoided. Another advantage of this method is to estimate the vocal tract filter as one transfer function and hence no pitch detection is required, which may simplify the existing speech processing. It is believed that with the ARMA model obtained using this proposed mothed, the CELP algorithm can achieve a synthetic speech of high quality at very low bit rates.