3krqhph Uhfrjqlwlrq Dv D Khdulqj Dlg Lq Whohskrqh Frppxqlfdwlrq

This thesis investigates the idea of using phoneme recognition as a telephone communication aid aimed at hearing-impaired people. A laboratory ASR system has been developed with which on-line trials (user studies) have been conducted in order to investigate and evaluate the idea of using state-of-the-art phoneme recognition as a " hearing aid ". Results from the on-line trials indicate that a significant improvement in communication quality may be achieved when using the phoneme recognition aid. The trials show an average 76% decrease in misunderstood words and an average 21% increase in communication speed when using the phoneme recognition aid. I would like to thank my supervisors at TMH/KTH, Mats Blomberg and Kjell Elenius for their help and continuous encouragement during this work, and my supervisor Lars Borin at the Department of Linguistics, Uppsala University for valuable feedback on the report and for his input regarding measures of significance. I also want to thank Anders Torberger at Polycom AB for giving me the opportunity to work with an inspiring project. A huge thank you to Håkan Melin for his untiring readiness in helping me out with Atlas, HMM theory, HTK,…. Many thanks also to: Giampiero Salvi for valuable help with HTK; Karl-Erik Spens for introducing me to the Speech Tracking procedure; Martin Dahlquist for his help with programming the simulation of the hearing impairment, and for valuable hints on methodology; Markku Haapakorpi for helping me out with numerous hardware issues; Botond Pakus, Kåre Sjölander and Jonas Beskow for hints on everything from tcsh programming to speaker adaptation. Thanks also to the rest of the staff at TMH for making this an instructive and pleasant time. Finally, many thanks to all of you who volunteered as subjects in the on-line experiments. 3UHIDFH This work constitutes phase two in the CTT 1 project " HörStöd " – a collaboration between the Department of Speech, Music and Hearing (TMH) at KTH and Polycom Technologies AB. The first phase (Sahlin, forthcoming), which was carried out at TMH in the fall of 1999, dealt with readability of phoneme strings. In that work off-line experiments were conducted, investigating the possibility of using the output of a phoneme recognizer as an aid in interpreting distorted (low pass filtered) speech. It also involved the development of a spelling dictionary word recognizer. The work described in this thesis is a continuation of the phoneme recognition part of the first phase. …