We explore timbre recognition by analogy with unsupervised pattern discovery in speech.
We segment the audio signals for each music note into fragments and clustering them using SOM. The segmentation of the output map could be pre-trained timbre recognizers for sequential matching using DTW.