Enhancement of alaryngeal speech using spectral subtraction

The transcervical electrolarynx is of great help to persons who cannot use their natural voice production mechanism. The device is held against the neck, and the vibrations generated move up the vocal tract to produce useful speech. The presence of background noise, caused by leakage of the acoustic energy from the vibrator, degrades the resulting speech. After formulating a theoretical basis, modified spectral subtraction method is used for speech enhancement. The average magnitude spectrum of noise, obtained with lips closed in the training mode, is subtracted from the magnitude spectrum of the noisy speech and the signal is reconstructed using the original phase spectrum. It is observed that effective noise cancellation is obtained, if the noise estimation and subtraction is done using 2-pitch frames. The set of optimal values for processing parameters of the method are empirically obtained.