Kevin J.P. Woods
  • Home
  • Research
  • Code
  • Sounds
  • CV
  • About
  • Links
Attentive Tracking Stimulus - synthetic voices intertwined in feature space
You'll hear a cue, mixture, and probe. Did the cue and probe come from the same voice?
Our ability to do this is robust to speech-like discontinuities!
Attentive tracking demo page on the lab website (more demos, with explanations)
'Bouncing':  Voices in a pitch "X" bounce instead of crossing
An odd case where the auditory system heavily favors the less probable interpretation of an ambiguous stimulus. 
This even happens even when you insert a silent gap where the voices would collide! 
Mixtures of speech are much harder to parse in reverse
Mechanisms of segregation and selection seems to rely on asymmetric temporal dependencies
Sound replacement by envelope matching
Ordinary sentences replaced piecemeal, drawing from a steely dan album and recordings of cats.
The result is overlaid with the original sentence.
Human vocals in a piece of music replaced with animal vocalizations, by the same process (find more here)
Top-down word perception
A three-tone pattern morphs into a word and single tone, if you group it a certain way...
Failure of consonants to stream correctly following temporal reordering of speech
Toward the end of this clip, the /h/ sounds appear to come from a separate source.
Sine-wave speech works even when speech is time-reversed
Sine-wave speech, originally due to Remez and Rubin, is a set of 3 - 4 sine waves heard as speech only after exposure to the original speech.
Grouping the components differently allows them to be heard as speech-- a phenomenon which appears to depend on linguistic structure.

However, this effect also works when speech is time-reversed, suggesting that the mechanism learning to 'restore' the sine-wave speech is not particularly sensitive to temporal dependencies in speech.
The effect also works when speech is in a foreign language, suggesting that the mechanism learning to 'restore' the sine-wave speech is not particularly sensitive to language-specific knowledge. 
Resynthesizing skillful performances with audio-to-midi
(preserve timing and nuance while modifying timbre)

​Glenn Gould
Prelude in C major
​
​Goldberg Var. 15
​

Gavotte
​
French Overture, Bouree II
​
Concurrent Variations
Pairs of Goldberg variations, time-warped to fit and aligned   (context)
Var. 17+19
​​

Var.20+23
​​
Var.1+5
​​
© COPYRIGHT 2015. ALL RIGHTS RESERVED.
  • Home
  • Research
  • Code
  • Sounds
  • CV
  • About
  • Links