A Formant Tracking Lp Model for Speech Processing in Car/train Noise

Formant estimation becomes complicated in the presence of correlated background noise such as car and train noise as the spectrum of noise from revolving mechanical sources have their own spectral peaks that affect the number and positions of the observed peaks in noisy speech spectrum. This paper investigates the modeling and estimation of spectral parameters at formants of noisy speech in the presence of car and train noise. Formant estimation using twodimensional hidden Markov models (2D-HMM) is reviewed and employed to study the influence of noise on observations of formants. The first set of experimental results presented show the influence of car and train noise on the distribution and the estimates of the formant trajectories. Due to the shapes of the spectra of speech and car/train noise, the 1 formant is most affected by noise and the last formant is least affected. The effects of inclusion of formant features in speech recognition at different SNRs are presented. It is shown that formant features provide better performance at low SNRs compared to MFCC features. Finally, for robust estimation of noisy speech, a formant tracking method based on combination of LP-spectral subtraction and Kalman filter is presented. Average formant tracking errors at different SNRs are computed and the results show that after noise reduction the formant tracking errors of 1 formant are reduced by 60%. The de-noised formant tracking LP models can be used for recognition and/or enhancement of noisy speech. This paper concentrates on formant estimation for speech in noisy car/train environments. In section 2 the extraction of formant features and the probabilistic modeling of variation of formant features are reviewed. Section 3 investigates the effect of noise on formant models. In section 4 a method, for robust estimation of formants of noisy speech, based on a combination of LP-spectral subtraction and Kalman filter is described. Section 5 concludes the paper. 2. FORMANT ESIMATION IN CLEAN SPEECH 2.1 Formant Track LP Models Speech is modeled as ) ( ) ( ) ( ) ( z L z V z G z X = (1) where G(z), V(z) and L(z) are the z-transform of glottal pulse, vocal tract and lip radiation[3]. The vocal tract is modeled by a cascade of second order formant-tracking LP models as