830 BELL SYSTEM TECHNICAL JOURNAL 



as fully as possible those parts of the problem that can be evaluated. 

 Such a relation involves two questions, (a) how do the articulations 

 of the sounds as measured with the testing lists compare with their 

 articulations as they are used in speech, (6) how should the articulation 

 values be weighted in order to obtain an index of the speech capabilities 

 of a system. 



In the first place, certain fundamental sounds of speech were 

 omitted from the above lists. The most important of these are the 

 consonant compounds. The majority of these sounds may be regarded 

 as the product of a very few combining consonants acting as modifiers 

 to the rest of the consonant alphabet. Since the combining consonants 

 or modifiers occur over and over in combination with various conso- 

 nants, it might be expected that the interpretation of the compounds 

 would depend primarily upon the interpretation of the various conso- 

 nants, and not upon the modifiers. In other words, the compounds 

 would be interpreted as simple consonant sounds. The tests discussed 

 below show that this is true on the average, although notable exceptions 

 may occur in individual cases. 



The testing lists were made up from the sounds shown in Table VI. 



TABLE VI 



Consonants Voivels 



Initial Final 



b, br, rb, b, a' 



d, dr, rd, d, a 



e 

 i 

 o 



These sounds were combined at random into syllables of the con- 

 vow-con-con and con-con-vow-con form. Ten lists of 90 syllables 

 each were made, and a crew of 10 callers, with 5 observers per caller, 

 was used. With this number of tests the probable error in the per 

 cent articulation for each sound is approximately 5 per cent. The 

 tests were made on the auxiliary circuit of the master reference 

 system. The sensation level of the received speech was about 80 db. 

 The results are shown in Table VII. 



