Intelligibility of time-reversed speech

Time-reversed speech has the same long-term amplitude spectrum as natural speech, but is largely unintelligible.
However, Saberi & Perrott ("Cognitive restoration of reversed speech," Nature 398: 760, 1999) recently showed that
local time reversal does not destroy intelligibility, if the time reversal window is brief. They divided everyday English
sentences into segments of fixed duration, and each segment was time reversed. Near-perfect intelligibility was
found for windows less than 50 ms and performance declined progessively as the time reversal window length was
increased to 300 ms.

The effects of local time reversal are illustrated with audio examples below. The example is of a male voice saying
"The watchdog gave a warning growl." In each case the waveform of the sentence was divided into successive,
non-overlapping time-reversal windows of progressively longer duration: 0 ms (unaltered original),
10 ms, 20 ms, 50 ms, 100 ms, 250 ms, 500 ms, and in the final example, 1 second. The figure shows
the waveform of a time-reversed sentence using a 0-ms time reversal window (unaltered original, top);
a 1-s time reversal window (middle); and a 50-ms window (bottom).

.WAV format .AU format

0-ms window (original)                0-ms window (see waveform, upper trace)

10-ms window                           10-ms window

20-ms window                           20-ms window

50-ms window                           50-ms window (see waveform, bottom)

100-ms window                         100-ms window

250-ms window                         250-ms window

500-ms window                         500-ms window

1-s window                               1-s window (see waveform, middle)

Other projects in speech perception

      Email: assmann@utdallas.edu
      Peter Assmann
      School of Human Development
      The University of Texas at Dallas, Richardson, Texas, 75083, USA
      Tel: (972) 883-2435
      FAX: (972) 883-2491