Enhancement of electrolarynx speech based on perceptual weighting

The electrolarynx plays an important role in phonation for the laryngectomees, however, the resulting speech is of less intelligibility and poor audibility since the present of a steady and colored additive noise. This paper, therefore, investigates the problem of the electrolarynx speech enhancement by taking into account the frequency-domain masking properties of the human auditory system and reduces the perceptual effect of the residual noise. Considering the particular characteristics of electrolarynx speech, the perceptual weighting technique is developed and incorporated into the traditional spectral subtraction algorithm to shape the residual noise and make it inaudible. The results suggest that the background noise can be reduced efficiently while the distortion of electrolarynx speech remains acceptable, the enhanced speech also sounds more pleasant to human listeners, suggesting that the proposed algorithm achieved a better performances of noise reduction over other subtractive-type algorithms.