ARTICULATION TESTING METHODS 



841 



the capabilities of systems to transfer thought by means of speech. 

 For giving a complete picture, it is necessary to give the articulation 

 values for each speech sound. Since this involves 36 articulation 

 values, it is difficult to compare various systems. To combine these 

 values into averages raises the question of how such an average shall 

 be taken. At first thought it might seem obvious that the weights 

 assigned to each sound articulation value should be proportional to 

 the frequency of occurrence of that sound in English speech. Many 

 of the most frequently occurring words, however, such as the, of, 

 and, to, in, that, etc., do not carry much of the thought, so that it 

 seems reasonable to exclude the effects of such words in the weighting 

 process. It is evident that many sets of weighting factors could be 

 evolved depending upon how far the exclusion process is carried and 

 depending upon whether written or spoken English is used, in deter- 

 mining the frequencies of occurrence of the sounds. After excluding 

 the twenty or twenty-five most common words, however, further 

 exclusion does not appreciably change the calculated articulation 

 value. The table below gives a set of factors obtained from the 

 frequencies of occurrence used in Table VIII. They are based upon 

 the studies of Messrs. French and Koenig ^° on the frequencies of 

 occurrence of speech sounds in spoken English. The effects of the 

 more common parts of speech, such as, personal pronouns, definite 

 articles, conjunctions, and prepositions have been excluded. 



TABLE XIV 



It will be noticed that the speech sounds are arranged in five groups. 

 The sounds in each group have very similar characteristics, so instead 

 of dealing with 36 articulation values for a circuit, it is only necessary 



^^ "Frequency of Occurrence of Speech Sounds in Spoken English," N. R, French 

 & W. Koenig, Proc. Acoustical Society of America, 1929. 



