First International Conference on Spoken Language Processing (ICSLP 90)
A method of processing noisy speech to enhance spectral features corresponding to the vocal tract system is presented. A new method of extracting pitch from noisy speech data is also presented. These methods depend on processing the Fourier transform phase through group delay functions. The basic idea used in these methods is that the features of noise and an all-pole system are distinct in the group delay function. To enhance the spectral features of the vocal tract system a modified group delay function is derived by suppressing the features corresponding to noise from the standard group delay function. Similar ideas are used to extract the periodic component corresponding to pitch from the magnitude spectrum. Performance of these methods is demonstrated for noisy speech data.
Bibliographic reference. Yegnanarayana, B. / Murthy, Hema A. / Ramachandran, V. R. (1990): "Speech enhancement using group delay functions", In ICSLP-1990, 301-304.