Comprehension of Sounds Paper

For the last forty years scientists and engineers have been trying to create a computer-based speech recognizer which could be trained to recognize both a big amount of words and a variety of speakers, but so far with no significant success. The programs which have been designed so far are either bound to a single speaker and should be heavily trained (such as Siri by Apple), or are limited to a very small amount of words (e.g. some telephone directories which are only able to recognize the numbers). Those systems which are designed to recognize both different speakers and different words are heavy to manipulate and still produce a sufficient amount of mistakes and misunderstandings which makes them inefficient.So what is it about our human speech comprehension that makes it so unique and so difficult to reproduce in a machine? The aim of this paper is to look at the mechanisms of human phoneme comprehension and try to understand what phenomena affect our comprehension and make it so special.In this paper we are going to look at such phenomena as:• Speed of speech comprehension• Segmentation• Co-articulation• Categorical perception and lexical identification shift• Phoneme restoration effect