This clip is a very good example of auditory streaming. The sound was created by Roger Reynolds (now at UCSD) for a composition he wrote called "Archipeligo" [Reynolds83] and is used here with permission. The idea for this example came from Steven McAdams, whose thesis [McAdams84] talks about what needs to be done to a sound to make it sound like more than one auditory object [Bregman90].
We are excited about this example for two reasons. First, it is a very good example of auditory streaming. A single sound, an oboe in this case, is first heard as a single instrument and then splits into two sounds, a clarinet and a soprano. It is a very good demonstration of what it takes for sounds to separate and to sound isolated. The second reason we are excited about this clip is that it shows the power of the correlogram. We think the correlogram is a very rich representation and clearly shows the two sounds as distinct visual objects.
The cynics in our audiences have pointed out that we have just translated one impossible problem (auditory grouping) into another impossible problem (visual understanding). This is correct but we feel that if we can see the auditory objects then it might be possible to design algorithms that can do the same. This is a current research direction and one possible approach is described in Clip 5.
In this demonstration, the pitch of the two sets of harmonics is at first fixed. This is the original oboe sound. Then, after a few seconds the pitches of each set of harmonics are independently jittered and the sound separates into two components. The odd harmonics sound like a clarinet since a clarinet has mostly odd harmonics. The even harmonics are now an octave higher (since the fundamental is now twice as high as before) and sound like a soprano. An important reason that the harmonics sound like a realistic clarinet and soprano is that the vibrato rate (5-6Hz) is very natural.
|
Frequency
|
||
|
|
||
| Time Delay |