Alaryngeal Speech Enhancement: Towards a More Natural Speech
cancer has stereotypically been associated with mostly geriatric patients who smoke.
However, this ailment affects younger people as well and there are several other factors
that also cause laryngeal cancer such as voice overuse or prolonged GERD
(gastroesophageal reflux disease). Laryngeal cancer inevitably leads to surgery or
laryngectomy and thereafter, loss of phonation ... read more(i.e. loss of voice) by the patients. In
addition, vocal fold paralysis could also lead to loss of phonation. One
commonly-adopted method for voice rehabilitation after the laryngectomy is the use of an
electronic larynx device (hereafter referred to as the electrolarynx.) These devices
create a machine-like raspy and monotonous sound, which lacks speech clarity. However,
given the relatively small number of patients affected by this rare ailment, companies
consider the market insignificant and therefore do not invest in research and
development to improve this speech technology for the affected population, even though
the need is real. The main objective of this project is to research the gap between the
usual laryngeal speech and the speech produced by the present-day electrolarynx, and to
identify the major causes of the discrepancies. This effort would help to support the
creation of a better voice prosthesis for laryngectomees (people who have undergone
laryngectomy), especially given the dearth of research in this field. To achieve this
goal, some of the prior arts were researched and analyzed, weighing their merits and
re-evaluating some of the ideas to identify the most promising directions. Previous
works focused heavily on investigating new methods for noise reduction, which are
neither applicable to real-time enhancement of electrolaryngeal speech nor to the
removal of the mechanical perceptibility. Our approach looked at the problem from
various perspectives and proposes a novel Human Audio Sensing approach for the
electrolaryngeal speech enhancement. As an alternative for voice rehabilitation, an
articulatory pattern recognition method was proposed for speech. In this work, an
existing electrolarynx (Servox model) was studied, analyzed in detail and modified in
order to monitor the improvements. Given the studies and the results obtained, it was
concluded that implementing improvements to the present electrolarynx requires a
holistic approach. This approach addresses several factors, such (1) an overhaul of the
excitation signal or source sound, which comprises attention to the driving waveform and
prosody integration and (2) an efficient coupling or transmission of the generated ideal
excitation signal. This investigation also led to the design of a new product
incorporating breath sensors to help facilitate the production of a more
natural-sounding speech, which also eliminates the normal electrolarynx control buttons
thus paving the way for a hand-free product. In addition, another alaryngeal speech
production technique employing an articulatory recognition system was
At the request of the author, this graduate work is not available to view in the Tufts Digital Library until June 1, 2018.
Thesis (M.S.)--Tufts University, 2016.
Submitted to the Dept. of Electrical Engineering.
Advisor: Karen Panetta.
Committee: Paul Lehrman, and Ronald Lasser.
Keywords: Electrical engineering, and Acoustics.read less
- Component ID:
- To Cite:
- TARC Citation Guide EndNote