Enhancement of esophagus speech using harmonic plus noise model

Abstract

Patients whose voice boxes have been removed use either artificial larynx or esophagus as a source of excitation to the vocal tract. This excitation is shaped by the vocal tract to produce the speech output. Artificial larynx is a costly device and continuous use of batteries increases the recurring expenses. Further, the presence of background noise degrades the speech qualify. Production of esophagus speech is more convenient and advantageous. The production of speech using esophagus is cheap, as no extra device is needed. Also the hands are free while speaking for doing some other work. But main problem of the esophagus speech is the very low amplitude of the output speech, intelligibility, and naturalness. The objective of this paper is to enhance the intelligibility of esophagus speech by using harmonic plus noise model (HNM). Esophagus speech is analyzed and synthesized using HNM and informal listening tests are conducted for accessing the improvement in the synthesized speech. Investigations show that the output is more natural and intelligible as compared to input speech signal but there is an increase in the random noise. This noise is uncorrelated to the input and can be suppressed by conventional techniques such as spectral subtraction method. Experiments are also carried out for enhancing the output of HNM by using Klatt synthesizer. Results show that the output of Klatt is almost clear as compared to the HNM output.