Contrastive study on the enhancement of pitch adjustable electrolarynx speech based on spectral subtraction technology

Spectral subtraction method is the most common way to eliminate the radiated noise of Electrolarynx (EL) whose pitch is constant. However, for pitch adjustable EL speech, it is not mentioned whether this method is effective. Thus, this paper conducts a study on the enhancement of pitch adjustable EL speech based on spectral subtraction technology. Several spectral subtraction algorithms including the Classical Spectral Subtraction (CSS), Multi-Band Spectral Subtraction (MBSS), Perceptual Weighting Spectral Subtraction (PWSS) and Weighting Function Spectral Subtraction (WFSS) had been performed and subjective and objective methods were used to evaluate and compare the performance. A native male speaker of mandarin Chinese who had been trained to be familiar with using EL was instructed to read twenty daily mandarin sentences in a soundproof room, which was recorded as the raw data of the EL speech. The subjective and objective results showed that the enhanced EL speech by CSS still remained noise obviously though part of radiated noise was eliminated. While the other three methods including MBSS, PWSS and WFSS could remove the noise effectively and improved the speech acceptability, but had little influence on speech Intelligibility. Therefore, the spectral subtraction method can upgrade acceptability of pitch adjustable EL speech and improve speech quality.