Applied Speech Enhancement for Personal Communication

Abstract: In our daily life noise is almost constantly present. At the same time our need to communicate reliably has probably never been this great. Surrounding noise impairs our ability to communicate by speech and causes the users of a speech communication system to strain both their hearing and their voices. This licentiate thesis deals with speech enhancement for personal communication. A method for speech enhancement mainly aiming at every day communication situations is presented. The method boosts speech energy in a communication system, leaving background noise unaffected and it also tracks changes in background noise characteristics. Hence it is an \emph{adaptive} speech enhancement algorithm. This is desirable since the noise characteristics of most every day noisy situations are more or less rapidly changing. Methods for facilitating personal communication in severely disturbed environments are also presented. Instead of placing the communication microphone in front of the mouth, the microphone is placed inside the external auditory canal of the user. A pair of ear-muffs equipped with an active noise cancelling (ANC) system are fitted onto the user's head. This setup, possibly combined with some speech enhancement method, enables the user to communicate even under the most extreme noise situations. The setup characteristics are examined and the resulting speech quality and intelligibility are evaluated using a speech recognizer based on Hidden Markov Models.