Electronic speech recognition

Abstract

The field of electronic speech recognition is only just beginning to be explored. Automatic recognition of the spoken word can serve many useful :purposes. Commands can be spoken into a computer or mechanisms controlled by voice inputs. Telephone :numbers could also be requested verbally from a suitably equipped exchange. At present some use is being made of the automatic recognition of digits to help in such tasks as the sorting of parcels and stocktaking where it is desired to keep the operators' hands free. However, all existing recognising machines are not completely accurate, the best giving accuracy of 90 - 95%. Speech recognition is one specialised form of the pattern recognition. However, the variability of the input is greater than in many other pattern recognising applications in that nominally identical utterances spoken by different speakers, or by the same speaker at different times are widely different. Conversely the total number of patterns to be distinguished is seldom more than about fifty. The main task in this project was to build a coder which reduced speech to certain quantised parameters. A combination of digital and analogue circuits were used. Speech recognition is in general a very difficult subject. Literature in this field is practically non-existant. This in fact was the main difficulty in this project; because of this there was no knowing whether a particular approach being taken was the correct one. In fact quite a few promising approaches were tackled which eventually led to no useful results which could be used in speech recognition. [...]