Effect of pole/zero manipulation in estimating the group delay spectrum

Acoustical science and technology

article

Husne Ara Chowdhury¹  M. Shahidur Rahman¹ 

[1] Department of Computer Science and Engineering, Shahjalal University of Science and Technology

关键词: Magnitude spectrum;    Phase spectrum;    Group delay spectrum;    Spikiness;    Poles/Zeros;   

DOI  :  10.1250/ast.43.232

学科分类：声学和超声波

来源: Acoustical Society of Japan

【 摘 要 】

Most speech signal analysis tasks ignore the use of phase spectrum and usually exploit the magnitude spectrum. However, some studies showed the importance of phase spectrum to preserve speech quality and intelligibility. The group delay (GD) spectrum is the alternative and meaningful representation of the phase spectrum. The negative derivative of the phase spectrum is called the GD spectrum. Researchers [1–4] use the GD spectrum to extract the basic features for speech recognition, synthesis, coding, voicing detection, etc. The GD spectrum resembles the magnitude spectrum and related to it through cepstral coefficients [1]. Thus it is reasonable to extract features from the GD spectrum, likewise the magnitude spectrum. Some attractive properties of the GD spectrum [2] make it a high-resolution spectrum, which becomes good enough for estimating the formant frequencies. Spikiness is the prime problem of the GD spectrum estimation. Bozkurt et al. in [3] estimated the GD spectrum from the magnitude spectrum and employed the analysis window other than the unit circle to avoid the spikes. However, this method shows lower performance for noisy speech owing to the noise effect on the poles/zeros near the unit circle.

【 授权许可】

Unknown