476 BELL SYSTEM TECHNICAL JOURNAL 



speech which are weak and submerged in the noise and therefore 

 contribute little or nothing to the intelligibility. Experiments have 

 shown the effect on voice transmission of removing portions of the 

 frequency range.^ Articulation tests were used to afford a quantitative 

 measure of the recognizability of received speech sounds. These show 

 that the upper frequencies may be cut off down to about 3000 cycles 

 without serious reduction in articulation. After such treatment, as the 

 noise level increases, the weaker and less articulate sounds become more 

 and more submerged in the noise and additional reduction in the 

 detrimental effect of the noise is required. 



In addition to the speech waves covering a frequency band they 

 occupy intervals of time. The unoccupied intervals between the 

 speech sounds contain noise. Reduction of the noise reaching the ear 

 in these intervals has been found to result, under certain conditions, in 

 an improvement in speech reception. This may possibly be explained 

 by considering the characteristics of the ear.^ It has been shown that 

 noise present at the ear has the effect of shifting the threshold for 

 hearing other sounds or has a deafening effect. That is, there is a 

 reduction of the capacity of the ear to sense sounds in the presence of 

 noise. For example, if a person has been listening to a noise for a 

 certain period, his ear is made insensitive so that speech signals 

 following are not so easily distinguished. The ear has a sensory build- 

 up time, that is, a time needed for the noise to build up to a steady 

 loudness. By reducing the noise in the intervals of no speech the 

 average threshold shift seems to be diminished. Aside from this the 

 presence of the noise tends to distract the attention from the perception 

 of the speech. Removal of noise during the intervals of no speech 

 tends to reduce this effect. 



Requirements 



In considering the elimination of the noise during these intervals it is 

 necessary to bear in mind certain characteristics of speech. ^ Speech 

 waves may be regarded as nonperiodic in that they start at some time, 

 take on some finite values and then approximate zero again. In 

 connected speech it is usually possible to approximately distinguish 

 between sounds and to ascribe to each an initial period of growth, an 

 intermediate period which in some cases approximates a steady state 

 and then a final period of decay. The duration intervals of various 

 syllabic sounds vary from about .03 to as much as .3 or .35 second. 

 When noise is high the weaker initial and final sounds become obscured 

 so that they contribute little to the intelligibility. 



^ See end of paper for references. 



