Sound Demo
 

This page demonstrates a sound separation method based on binaural cues extracted from the responses of a KEMAR dummy head. The system was systematically tested for multiple source configurations in anechoic conditions.

For more details see:

N. Roman, D. L. Wang and G. J. Brown, "Speech segregation based on sound localization," J. Acoust. Soc. Am., vol. 114, pp. 2236-2252, 2003.


Two-source configuration (Target at 0o ; Noise at 30o)

 Target 
 
    Noise    Mixture    Reconstructed target 
Pure Tone
White Noise
Noise Burst
Cocktail Party
Rock Music
Siren
    Trill Telephone 
Female Speech
Male Speech
Female Speech

 Three-source configuration (Target at 0o; Noise 1 at -30o; Noise 2 at 30o)

 Target    Noise 1 
 
    Noise 2   Mixture    Reconstructed target 
Pure Tone
White Noise
Noise Burst
Cocktail Party
Rock Music
Siren
    Trill Telephone 
Female Speech
Male Speech
Female Speech

Speech signals of the training set
 
 
  ID    Speaker ID  Utterance   Wave Sound
S0 MKLSO "Primitive tribes have an upbeat attitude"
S1 FCKE0 "Only the best players enjoy popularity"
S2 MCDC0   "Our aim must be to learn as much as to teach" 
S3 FEAR0 "Development requires a long-term approach"
S4 FDMS0 "Poets, moreover, dwell on human passions"
S5 FETB0 "Change involves the displacement of form"
S6 FCMM0   "The system works as an impersonal mechanism" 
S7 MJWS0 "Most assuredly ideas are invaluable"
S8 MRVG0 "False ideas surfeit another sector of our life" 
S9 MJRH0 "But in every period it has been humanism"