6. Results
Table I shows the recognition rates for various numbers of instruments and two different analysis lengths. The
number of windows 3 and 20 corresponds to approximately 93 ms and 490 ms, respectively.
Number of instruments
Number of windows 3 5 8 10 15 39
3 98.3 92.9 88.7 81.5 74.4 54.3
20 93.9 92.6 88.0 86.5 81.6 63.6
Table I. The recognition rate (%) for a various number of instruments and two analysis lengths.
Compared to the previous experiment using only the steady-state portion of the sounds, the current system
achieves 10-20% increase in the recognition rate. For example, the recognition rate of the trumpet, the clarinet,
and the violin using the dynamic spectra is 98% compared to 80% using the steady-state spectrum and the
recognition rate involving 39 timbre improved from 50% to 64%. These results are comparable to those reported
by Martin and Kim (1998) who also used the McGill CDs. Note that these results are considerably better than
reported human performances (Kendall 1986, Saldanha and Corso 1964).
In typical ensemble situations, one is usually dealing with handful of instruments. Results here show that a reliable
real-time system for recognizing instruments within the attack portion (less than 100 ms) is possible.
7. Conclusions
Using exemplar-based learning model, the computer was able to identify orchestral instruments quite accurately
and quite quickly, especially when a small number of instruments were involved. These results indicate the
feasibility of using timbre recognition tasks in real-time and in real-life applications. The obvious next step in this
research is to experiment with other sources of timbre.
Bibliography
Cover, T., and P. Hart. 1967. Nearest neighbor pattern classification. IEEE Transactions on Information Theory 13
(1): 21-7.
Fujinaga, I. 1998. Machine recognition of timbre using steady-state tone of acoustic musical instruments.
Proceedings of the International Computer Music Conference. 207-10.
Holland, J. H. 1975. Adaptation in natural and artificial systems. Ann Arbor: U. of Michigan Press.
Kendall, R. A. 1986. The role of acoustic signal partitions in listener categorization of musical phrases. Music
Perception. 4 (2): 185-214.
Martin, K. D., and Y. E. I(im. 1998. Musical instrument identification: A pattern-recognition approach. Paper read
at the 136' meeting of the Acoustical Society of America.
Nosofsky, R. M. 1986. Attention, similarity, and the identification categorization relationship. Journal of
Experimental Psychology: General 115 (1): 39-57.
Nosofsky, R. M. 1984. Choice, similarity, and the context theory of classification. Journal of Experimental
Psychology: Learning, Memory, and Cognition 10 (1): 104-14.
Puckette, M. S., T. Apel, and D. D. Zicarelli. 1998. Real-time audio analysis tools for Pd and MSP. Proceedings
of the International Computer Music Conference. 109-12.
Saldanha, E. L., and J. F. Corso. 1964. Timbre cues and the identification of musical instruments. Journal of the
Acoustical Society of America. 36(11): 2021-6.
Wettschereck, D., D. W. Aha, and T. Mohri. 1997. A review and empirical evaluation of feature weighting
methods for a class of lazy learning algorithms. Artificial Intelligence Review 11 (1-5): 272-314.
ICMC Proceedings 1999
-177 -
0