FKEQUENCY SELECTIVITY 



245 



ulation. The curve for low-pass filters can 

 be similarly interpreted. 



The two curves of Fig. 1 cross at 1900 cps. 

 This value is obtained for men's and women's 

 voices. For men's voices alone the crossing 

 point is at 1660 cps (1). The frequencies 

 above this crossing point contribute as much 

 to intelligibility as do the frequencies below. 

 If we define the range from 200 to 6000 cps 

 as contributing 1.00 to intelligibility, then 

 200 to 1900 cps contributes 0.50, and 1900 

 to 6000 cps contributes 0.50, and the range 

 of speech frequencies is divided into two 

 bands of equal importance. It is possible 

 to repeat this process of subdivision until 

 the frequency limits of x bands, each contrib- 

 uting 1/x to intelligibility, are determined. 



This weighting of the frequency scale ac- 

 cording to importance for intelligibility en- 

 ables us to compare communication systems 

 with different frequency-response character- 

 istics. It is necessary, however, to weight 

 the contribution of each of the x bands ac- 

 cording to the signal-to-noise ratio in the 

 band. A band begins to contribute to in- 

 telligibility when it is only slightly above the 

 threshold of detectability, but does not make 

 its maximum contribution until it is 50 db 

 above the threshold. 



It is possible, by weighting the frequency 

 and intensity scales in this manner, to pre- 

 dict the adequacy of a communication sys- 

 tem from a knowledge of its frequency-re- 

 sponse and amplification, the noise level, and 

 the type of speech materials (1, 2, 3, 4, 5, 12 

 24). Such computational procedures are a 

 great convenience in the design of commun- 

 ication systems, and make it possible to min- 

 imize the amount of articulation testing re- 

 quired. This does not mean, however, that 

 articulation testing methods can be aban- 

 doned. The computations provide a first 

 approximation, but their accuracy must be 

 increased. Many of the things which can 

 go wrong in a communication system are 

 not considered. Much further research is 

 needed before we can be satisfied with this 

 theoretical interpretation of the perception 



of speech. An effort should be made to 

 maintain the scientists and facilities for such 

 experimentation . 



Sound-powered telephones provide an ex- 

 ample of the practical applications of this 

 research. The output level of a sound-pow- 

 ered system can be increased if the frequency 

 range passed by the system is reduced. Is 

 it better to have a wide-band system with 

 a weak output, or a narrow-band system 

 with a loud but distorted output? The in- 

 formation necessary to answer this question 

 is summarized by the equal-articulation con- 

 tours in Fig. 2 (9). These curves were de- 

 termined with a variety of band-pass systems 



FREQUENCr ifJ CYCLES PER 



Fig. 2. Equal-articulation contours for band- 

 pass systems in noise. The parameter is the 

 articulation score. The abscissa indicates the 

 lower and upper cut-off frequencies of the band. 

 The ordinate gives the increased amplification 

 necessary for the narrow-band system to maintain 

 the same articulation score as a wide-band system. 

 Men's voices. (After Egan and Wiener, 9) 



tested with nonsense syllables in the pres- 

 ence ol uniform white noise. On the ordi- 

 nate, zero db is the intensity of the speech 

 in a wide-band system. In order to main- 

 tain any given articulation score, the level 

 of the speech in a narrower band must be 

 increased. The curves are used in the fol- 

 lowing way. Assume that a band-pass sys- 

 tem extending from 950 to 2400 cps is to be 

 widened to 550 to 4200 cps. In order to 

 maintain an articulation of 10 percent, how 

 much loss in level can be tolerated in the 

 wider band? Fig. 2 indicates that the artic- 

 ulation score of 10 percent could be main- 

 tained even though the level of the received 

 speech provided by the wider band was 9 



