348 BELL SYSTEM TECHNICAL JOURNAL 



Methods of Control and Analysis 

 The Articulation Testing Method 



The articulation testing method itself remains essentially as it has 

 been described previously in this journal.^ Briefly a test is carried out 

 as follows: Each of a number of persons, referred to as callers, speaks 

 a list of meaningless monosyllables of the consonant-vowel-consonant 

 type over the circuit tested. The test syllables are spoken as part of a 

 sentence by inserting them at the time of calling in blank spaces left 

 for that purpose in the middle of a number of short sentences called 

 "carrier sentences"; as an example we have the sentence "When will 

 mid be done," the test syllable being nud. A group of observers at the 

 receiving end of the circuit record their understanding of the test 

 syllable; their records are compared with the syllable called and any 

 errors are noted. The results of the test may be expressed in various 

 ways: Sound articulation, meaning the percentage of sounds correctly 

 understood, (or its complement, sound errorj is generally preferred. 



The obtaining of a useful numerical index expressing the articulation 

 performance of a telephone system is made difficult by the combined 

 effect of a large number of variable factors: The sounds of speech are 

 numerous and subtle, covering a wide range both in frequency and 

 volume. Voices differ from each other and the same voice may vary 

 considerably in its characteristics from time to time. Hearing abilities 

 differ also, both among individuals and for the same person at different 

 times. Finally, attentiveness and skill in perception vary greatly. 



Because of these variable factors the attainment of precision, in the 

 sense of the closeness with which a numerical result can be reproduced, 

 depends upon the careful formulation of the technique and incessant 

 watchfulness in supervision. Precision, however, is not enough. Two 

 persons, one acting as caller, the other as observer, might, with training, 

 learn to reproduce their experimental data closely but with very little 

 practical value. It is essential that the results be representative of a 

 large number of persons, as well as precise. Both men and women 

 should be represented, for example, because distortion at high fre- 

 quencies affects the reproduction of their voices quite differently. In 

 the present testing group four men and four women serve as callers, 

 each calling a list of 66 syllables to four observers; this constitutes a 

 single test. In the paper referred to above a detailed discussion is 

 given of the precautions which it is necessary to take in the selection 

 of voices, the testing of the observers' hearing, the training of the crew, 

 and the formulation of the lists of syllables used in testing. 



' "Articulation Testing Methods," H. Fletcher and J. C. Steinberg, B. S. T. J., 

 VIIJ, p. 806, October, 1929. 



