AM-FM decomposition of speech signal using MWL criterion

Adaptive maximum windowed likelihood algorithm is introduced and adapted to decompose the speech signal as an amplitude modulated-frequency modulated (AM-FM) signal. Choosing the window length and type, the algorithm can be adjusted to decompose different pieces of speech signal. By tuning the step size for each frequency, the algorithm can be tuned for each formant frequency. Simulations for two phonemes and an all voiced piece of speech show that the algorithm is able to track the formant frequencies successfully unless it is used in highly changing formants where some treatments have been suggested.

[1]  Peter C. Doerschuk,et al.  Statistical AM-FM models, extended Kalman filter demodulation, Cramer-Rao bounds, and speech analysis , 2000, IEEE Trans. Signal Process..

[2]  Shan Lu,et al.  Nonlinear modeling and processing of speech based on sums of AM-FM formant models , 1996, IEEE Trans. Signal Process..

[3]  S. Gazor,et al.  Adaptive maximum windowed likelihood AM-FM signal decomposition , 2003, Proceedings of the 3rd IEEE International Symposium on Signal Processing and Information Technology (IEEE Cat. No.03EX795).

[4]  Ramdas Kumaresan,et al.  On separating voiced-speech into its components , 1993, Proceedings of 27th Asilomar Conference on Signals, Systems and Computers.

[5]  Simon Haykin.  First ASSP Workshop on spectral estimation , 1981 .

[6]  Jerry D. Gibson,et al.  Digital coding of waveforms: Principles and applications to speech and video , 1985, Proceedings of the IEEE.