Investigations of the Effect of Nonlinearly Generated Excitations on the Quality of the Synthesized Alaryngeal Speech

Objectives: To explore the use of ultrasonic waves based glottal excitation for reducing the distracting background noise present in the conventional artificial larynx. Methods/Analysis: The persons, whose larynx is removed, mostly use artificial larynx to generate the excitation, which produces background noise. Our hypothesis is that if the glottal excitation is generated by using two ultrasonic waves, the background noise may be reduced as the ultrasonic frequencies are inaudible. The audible excitation is generated within the vocal tract by nonlinear interaction of the external high frequencies focused around the larynx. Vocal tract parameters were extracted from the recorded normal and alaryngeal speech segments using Linear Predictive Coding (LPC) technique using analysis window of 20 ms with overlap of 10 ms and order of 500. Findings: Five types of excitations, sinusoid of the desired pitch frequency, sinusoid of high frequency, sinusoid of high frequency plus pitch, amplitude modulated high frequency and non-linearly generated excitation are used. The analysis of the results shows that non-linearly generated excitation may be used for reducing the background noise along with enhanced intelligibility and Mean Opinion Score (MOS) estimated for the cardinal vowels using non-linear excitation is 2.1, 4.0 and 3.7 with laryngeal vocal tract parameters, for the three cardinal vowels as compared to the MOS of 1.4, 1.0 and 2.3 with alaryngeal vocal tract parameters. Novelty: In the proposed technique, use of ultrasonic frequencies for generating the glottal excitation reduces the distracting audible background noise present in conventional techniques resulting in enhanced alaryngeal speech.