Timbre Similarity

Presentation

Summary

Annotated Bibliography

• Brown, J. 1999. Computer identification of musical instruments using pattern recognition with cepstral coefficients as features. Journal of the Acoustical Society of America. 105: 1933–41.

  Paper by a physicist - not in musical circles. Used 1 single long recording to train sax and oboe sounds. Continually retried which sample to pick on an ad hoc basis. Used Gaussian classifier to classify set.

• Fujinaga, I. 1998. Machine recognition of timbre using steady-state tone of acoustic musical instruments. Proceedings of the International Computer Music Conference. 207–10.

  Utilizes hand picked steady state segments of McGill instrument samples. Uses GA and KNN to classify.

• Fujinaga, I. and K. MacMillan. 2000. Realtime recognition of orchestral instruments. Proceedings of the International Computer Music Conference. 141–3.

  Realtime extension of Fujinaga's 1998 paper. Realtime. Utilizes PD for analysis and GA+KNN for classification.

• Grey, J., and G. Gordon. 1978. Perceptual effects of spectral modifications on musical timbres. Journal of the Acoustical Society of America. 63(5): 1493-1500.

  First paper on timbre similarity. Utilized pairwise similarity ratings of 16 tones and their simplified versions. Used MDS to conclude that timbre has 3 dimensions. Accounts spectral centroid as 1 dimension.

• Herrera, P., A. Yeterian, and F. Gouyon. 2002. Automatic classification of drum sounds: A comparison of feature selection methods and classification techniques. International Conference on Music and Artificial Intelligence. 2: 69-80.

  Used sample of 634 drum samples from different sample cds. Split features between attack time and decay time. Only MFCC and empirically derived energy bands cover the entire signal.

• Tindale, A., A. Kapur, G. Tzanetakis, and I. Fujinaga. 2004. Retrieval of percussion gestures using timbre classification techniques. Proceedings of the International Conference on Music Information Retrieval. 541-5.

  Classificaion of different types of snares. Used 3 different classifiers - KNN, SVM, and Neural Nets.