Can perceptual filling-in of speech weaken an accent?

by Elian Malkin

The picket-fence effect with speech, discussed in Perceptual Restoration of Missing Speech Sounds (Warren), is an example of perceptual filling-in. The auditory system in some cases perceives discontinuous sounds as continuous when discontinuity is masked with noise. The simplest example of this illusion can be experienced by listening to a pure tone with a gap filled by masking noise, which tricks the auditory system into perceiving the pure tone as continuing through the masking noise. In the speech illusion, a sample of speech is muted at regular intervals, and masking noise is likewise placed in the gaps. If the gaps are short enough, the speech will appear continuous behind the noise. The masking noise must be loud enough and contain the appropriate frequencies to have potentially masked the speech, so that the auditory system can infer continuity.

I wanted to explore whether the filling-in of speech is influenced by heavily accented voice. My illusion presents a speech sample in which heavily accented segments are silenced, yet subtle accent cues still remain. I selected the segments to silence by researching the mispronunciations typical of my chosen accent, as well as by listening for particularly unusual pronunciations in the speech sample. The silenced segments are filled by pink noise - similar to white noise but with louder low frequencies. This noise effectively masks the male voice in the demo while sounding less harsh than white noise by being softer at the high frequencies. My hypothesis is that the auditory system will be biased towards more familiar accents, and will fill the masked gaps with unaccented speech. Try to see whether the accent in the unedited speech sample is stronger than expected - the parts filled in by your auditory system will now be the heavily accented original pronunciations. The first sound presents the discontinuous speech alone, followed by the speech with masking noise, and finally the unedited speech sample.

The speech may be difficult to understand, therefore I have provided a transcript below so that one can focus on whether one's auditory system indeed fills in a neutral accent:
"In the photoshoots you do with, uh, Vogue magazine and Cosmopolitan and, uh, entertainment weekly Time magazine, your hair changes dramatically, uh, from short to long. Do you wear wigs?"

PLEASE PLAY QUIETLY AT FIRST

Discontinuous Speech:

Masked Speech:

Unedited Speech:

Citations:

Warrren, Richard M. & others,. (1970). Perceptual Restoration of Missing Speech Sounds. Science. 167. 392-393. 10.1126/science.167.3917.392.

The audio file is from a really great prank interview with Elijah Wood. I highly recommend.

Comments

Elian Malkin
c) The subjective reports suggest that when accented speech is partially masked, the masked segments are perceptually filled in with a more familiar accent to the observer. The reports state that this change is subtle, therefore answering the question posed by indicating that perceptual filling-in of speech can slightly affect one's perception of an accent.

d) I wonder if priming the observer by first having them hear the unedited speech would completely eliminate the perceptual difference between the masked and unedited samples?

Katarina Bulovic
a) To me, the masked speech sounded very similar to the unedited speech, but I did perceive the masked parts of the speech slightly differently from what I heard in the complete recording.
b) As a researcher, this does support the hypothesis, because the masking of heavily accented portions of speech is filled in with speech that sounds different from the original recording.

Nicholas Guiliano
a) Comparing the masked speech with the unedited speech, I did in fact detect subtle changes in my perception of the speaker's accent. It wasn't a dramatic change, which potentially could be due to a partial familiarity with the accent. Interestingly, going back to the masked sample after hearing the unedited speech, the accent was easier to distinguish.

b.) The change in perception of the accent indicates that the hypothesis is correct- masking of the heavily accented portions is filled in with more familiar (i.e. differently accented) continuous sound.