Sound examples of percussion detection

These are examples of applying a probabilistic model-based method for detection and labeling of percussion instrument sounds. Examples include samples from two Turkish percussion instruments, the Darbuka and the Bendir. Isolated strokes of both instruments were recorded in anechoic and normal room conditions with two different strokes, düm (middle shot) and tek (rim shot). The recorded isolated sounds were used to train the model. Then, the algorithm was applied for detection and labeling of simple rhythmic patterns.

These examples have been constructed so that there is the original recording for reference and then a MIDI synthesis version of the same pattern, for which the onset times and instrument types were extracted from the detection data. These examples can be used as auditory evaluation of algorithm performance with different instruments in different conditions.

While comparing the original recording and the resynthesis version it should be noted that imperfect playing affects class labeling and that the original temporal structure is well preserved.

Bendir, anechoic conditions:

Düm-tek-tek pattern Original recording MIDI resynthesis

Bendir, normal room conditions:

Düm-tek-tek pattern Original recording MIDI resynthesis

Darbuka, anechoic conditions:

Düm-tek-tek pattern Original recording MIDI resynthesis

Darbuka, normal room conditions:

Düm-tek-tek pattern Original recording MIDI resynthesis