Our brains fill in the gaps in speech

Last week we wrote about using STIPA (Speech Transmission Index through Public Address) to test our Teraphone, the speech-optimised long-distance megaphone we have built for outdoor use. We believe STIPA gives a better indication of how well transmitted spoken messages can be received and understood than just measuring dB levels.



Initially we played the standard STIPA signal on an iPhone through a Wi-Fi network to a Bose Sound Dock and then measured the STI level using the iSTI App on an iPad.   At 0.98, the STI reading we observed was higher than the reading of 0.88 we got when playing the STIPA signal through the Teraphone and measuring at a distance of 10m.   The difference isn’t surprising given the differences in quality (and price!) of the speakers in the two systems tested.

Much more surprising was that when we moved far enough away from the Teraphone for the STIPA reading to come down to, say, 0.72, we still found that there was no difficulty at all in fully understanding a spoken message sent through the Teraphone!

The explanation lies in the amazing ability of the human brain to “fill in the gaps” when listening to imperfectly reproduced speech. This phonemic restoration effect was first reported by Richard Warren in 1970 (“Perceptual Restoration of Missing Speech Sounds”, Science, 167, 392-393, 1970). Warren showed that when an extraneous sound such as a cough or tone completely replaces a speech sound in a recorded sentence, listeners still believe that they hear the missing sound. The extraneous sound seems to occur during another portion of the sentence without interfering with the intelligibility of any phoneme.

The STI (which follows international standard IEC 60286-16 (2011)) takes this into account when it predicts the likelihood of syllables, words and sentences being comprehended:


You can see that even when the intelligibility of individual syllables is down to 45%, a native speaker will still make complete sense of around 90% of spoken messages.

 It is for this reason that a PA system is generally taken to be satisfactory when the STI value is greater than 0.5.