On Presenting the Output of a Mechanical Speech Recognizer
D. Β. Fry, Peter B. Denes · The Journal of the Acoustical Society of America · 1957
The output of a mechanical speech recognizer consists of a series of units which are closely correlated with the phoneme sequence which forms the speech input to the system. This output has to be transformed and presented to the human “reader,” usually in one of two ways, either by using it to reconstruct audible speech sequences or by presenting it in a sequence of visible characters. In both cases, the success of the recognizer will depend greatly on the degree to which good use can be made of the reader's existing linguistic knowledge, and methods of assessing its success must take this fact into account. This paper will present the results of experiments designed to compare methods of presenting the output and also methods of measuring the performance of the recognizer in terms of the ease with which the “reader” can reconstruct the input.