JAPAN-PROGRAM Archives

May 1999

JAPAN-PROGRAM@LIST.UVM.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Steve Cavrak <[log in to unmask]>
Reply To:
UVM Japan Program News and Discussion List <[log in to unmask]>
Date:
Mon, 3 May 1999 12:01:23 -0400
Content-Type:
text/plain
Parts/Attachments:
text/plain (128 lines)
/* The following article / online experiment was spotted
   by Virginia Clark, UVM Professor of Linguistics (Emeritus)
   and former chair of the English Department. I found it
   encouraging, and perhaps a partial explanation of why
   native Japanese speakers are generally so encouraging
   to non-native learners - to our ears we sound terrible,
   to their ears, it's just a heavy accent.  -- Steve
*/


Back Talk

Scrambled Words, Unscrambled Meaning

You can make out words even if they’re distorted or
almost drowned out by other noise. Now researchers show
you can even understand speech that is played back in
small time-reversed chunks. (ABCNEWS.com)

By Kenneth Chang
ABCNEWS.com

--------------------------------------------------------
http://www.abcnews.go.com/sections/science/DailyNews/speechreverse990428.html
--------------------------------------------------------


--------------------------------------------------------
SUMMARY : The ability of the human brain to understand
speech is robust, even when the words are badly distorted.

"This is a nice demonstration of a very important
property of spoken language, namely its resiliency
to all sorts of distortions that happen in every
day life." ... Steven Greenberg, International
Computer Science Institute.
--------------------------------------------------------


Divide a recording of a spoken sentence into segments,
each 1/20th of a second long. Flip each segment so that
it plays backward, then string the segments back
together in the correct order. Then play this string of
backward sounds. You might expect mangled noise.
Individual consonants and vowels are generally shorter
than 1/20th of a second, which means the word sounds
have been chopped apart and reassembled in the wrong
order. But we don’t always hear what we think we hear,
and that’s a good thing. Intuitively and automatically,
the speech-decoding part of the brain sifts through
noise and distortion and focuses on the portions it can
understand.

--------------------------------------------------------

        Try the experiment yourself:

        Here;s the sentence:
        http://play.rbn.com/?swave/abc/demand/990427original.ra

        Here's the mangled version of the same
        sentence, chopped and time-reversed in
        short chunks.
        http://play.rbn.com/?swave/abc/demand/990427saberi.ra

--------------------------------------------------------

Still Sounds Like English

“The first time, I listened to it, I was quite shocked,”
says Kourosh Saberi, a researcher in the California
Institute of Technology’s division of biology. “The
sentence is so clearly audible.” Except for clicking
noises where the sound segments were digitally spliced
together, the words sound basically the same. All seven
test subjects understood the sentences perfectly.
Saberi, along with David Perrott, a psychology
researcher at California State University, Los Angeles,
were exploring how much speech could be distorted and
still understood. The research, reported in Thursday’s
issue of the journal Nature, could help explain how we
focus on one conversation at a noisy party or comprehend
someone with a heavy accent. A better understanding of
how the brain processes speech could also lead to better
voice-recognition software.

More than Vowels and Consonants

“This is a nice demonstration of a very important
property of spoken language,” comments Steve Greenberg,
a scientist at the International Computer Science
Institute affiliated with University of California,
Berkeley, “namely its resiliency to all sorts of
distortion that happens in everyday life.” Greenberg has
championed the notion that the brain does not key on
individual word sounds to decipher speech, but instead
the meaning is encoded on slower, syllable-length
variations in sound patterns. He gives the analogy of a
movie. Swap around a couple of frames in the film, or
even a handful of frames, and you won’t notice anything
has been changed, because the brains blurs together the
images together to form the illusion of continuous
motion. Vowel and consonant sounds are analogous to
individual film frames, Greenberg says, and the
underlying meaning of speech is likewise resilient to
brief manglings. In Saberi and Perrot’s experiment, when
the sentence was chopped into longer time-reversed
chunks of about one-seventh of a second each, listeners
could make out what was being said only half of the
time. Time-reversed segments of 1/5th of a second or
longer were unintelligible gibberish.


Like an Accent

With the harder-to-understand examples, listening to the
same sentence over and over again sometimes helped. “To
us, this is some kind of cognitive recalibration,”
Saberi says, “similar to listening to a newly heard
accent. I had this trouble with my adviser at Berkeley.
For the first couple months I had the hardest time
understanding what he was saying. Not because of the
content, but the way he talked [fast and mumbling]. Now
I have no trouble understanding him.” And that likely
would be true even if his advisor was speaking backward
in 1/10th of a second chunks. ehT emas si ton eurt fo
nettirw sdrow.

ATOM RSS1 RSS2