Machine recognition of timbre using steady-state tone of acoustic musical instruments

Recent experiments indicate that steady-state portion of an acoustic musical instrument may be sufficient for timbre recognition. Here a computer-based classifier was used to recognize very short samples of steady-state tones. Gregory Sandell’s SHARC database consisted of 39 different timbre (23 orchestral instruments, some with different articulations) played at different pitches (total of 1338 spectra) were used as the samples for an exemplar-based learning system that incorporates k-nearest neighbor classifier with genetic algorithm. The latter is used to find the optimal set of weights for the features to improve the classification. The features calculated from the spectral data of the steady-state portion of the instrumental sound included centroid and other moments, such as skewness and kurtosis. As expected the centroid alone was the best single feature with a recognition rate of 20%, which is much better than chance (2.5%). The best results were obtained using seven features: the fundamental, the integral of the spectrum, the centroid, the standard deviation, the skewness, and the first two harmonic partials. What was surprising was that the recognition varied greatly between instruments. While the French horn and the muted trumpet were recognized over 90%, the recognition of other instruments, such as the cello with martele (18%), and the violin pizzicato (14%) were very poor. The average overall was 50.3%.