Interleaved melody identificationInterleaved melody identification
The identification of familiar melodies is a task which engages stored representations or 'schemas' (Bregman, 1990). However, melodies which are readily identifiable when presented in isolation can be much more difficult to recognise when interleaved -- that is, when notes from one melody alternate with notes from another (Dowling, 1973). Interleaved melodies are important in studies of auditory scene analysis since they can become much easier to recognise when tones from one melody are acoustically distinguished from tones from the other.
The extensive study of Hartmann & Johnson (1991) motivated the current demonstration (which uses a subset of both the melodies and the conditions used by them). They studied the identification of interleaved melodies whose elements differed in such factors as frequency range, ear of presentation, level etc. They were able to rank the effectiveness of 10 such factors in promoting melodic stream segregation.
Load the demonstration by typing 'intmel' at the MATLAB prompt. A dozen melodies are preloaded into the tool, so there is no need to load any further data [Hartmann & Johnson's paper lists many more melodies which we will get round to incorporating one day!].
The first step is to select which melodies you are familiar with (there is no point attempting to recognise unfamiliar tunes, although you can learn them on the fly if you like). The melodies menu (1) lists those loaded. Choosing any of these toggles its selection state. Initially, two melodies are indicated as being known. If you don't know these, unselect them. However, there must be at least two selected for the demonstration to be fully enabled, for obvious reasons.
When you are familiar with the melodies, you are ready to identify interleaved pairs. Pairs are delivered by pressing the A+B button (3). A spectrogram of the stimulus appears in the top panel (2). You may press this button as many times as you like. To hear the individual melodies, use buttons A and B (4). When you wish to hear another pair randomly-selected from the subset of melodies you have selected as familiar, press the next... button (5).
The default is to present the melodies in their base form, which means that they are identical in all key elements (except for the melody...). Hartmann & Johnson chose the base melodies to occupy similar frequency ranges, so identification is quite difficult in this condition. To examine the effect of potential cues to allow melody discrimination, use the options 6-9:
- frequency separation: a specified number of semitones can be added to one or other melodies to increase the mean frequency separation between the pair of melodies.
- relative level: use popup menu 7 to specify the relative level (in dB) between the two melodies.
- duration ratio: the duration of melody B tones relative to those melody A is specified by popup menu 8.
- ear of presentation: melodies can be send to different ears by selecting the dichotic radio button (9). This will not work correctly on platforms which do not support stereo output.
[Aside: in a future release, the radio buttons accompanying options 6-8 will be removed since they are redundant.]
The presentation rate (in tone pairs per second) defaults to 4, but can be changed via the preferences menu (10). Similarly, the tone duration can be altered from its default of 80 ms.
- Bregman (1990). Auditory Scene Analysis. MIT Press.
- Dowling (1973). Cognitive Psychology, 5, 322-337.
- Hartmann & Johnson (1991). Music Perception, 9(2), 155-184.
Produced by: Martin Cooke
Release date: June 22 1998
Permissions: This demonstration may be used and modified freely by anyone. It may be distributed in unmodified form.