The paper presents a method for evaluating similarity of temporal amplitude envelope of violin sounds. The experienced violinmakers are able to separate good sounding from bad sounding violins just by regarding the envelope of individual sounds. The contours of two-sided envelopes of individual open-string sounds are considered as images. Four uncorrelated visual descriptors are used to form a feature vector characterizing image shape. Individual distance measures are selected for each feature. The similar objects are grouped using k-means method. Violin sounds from AMATI database have been used in experiments.
展开▼