论文信息 - MODEL-BASED EVENT LABELING IN THE TRANSCRIPTION OF PERCUSSIVE AUDIO SIGNALS

MODEL-BASED EVENT LABELING IN THE TRANSCRIPTION OF PERCUSSIVE AUDIO SIGNALS

In thispaperwedescribeamethodfor thetranscriptionof percussive audiosignalswhich have beenperformedwith arbitrarynondrumsounds.Thesystemlocatessoundeventsfrom theinputsignal using an onsetdetector . Then a set of featuresis extracted from the onsettimes.Featurevectorsareclusteredandthe clustersareassignedwith labelswhich describethe rhythmic role of eachevent.For thelabeling,a novel methodis proposedwhich is basedon metrical(temporal)positionsof thesoundeventswithin themeasures. Thesystemis evaluatedusingmonophonicpercussive tracks consistingof non-drumsounds.In simulations,the systemachieved a total error rate of 33.7%. Demo signalsare available at URL:<http://www.cs.tut.fi/~paulus/demo/>.

Anssi Klapuri | Jouni Paulus

[1] Aniruddh D. Patel,et al. Acoustic and Perceptual Comparison of Speech and Drum Sounds in the North Indian Tabla Tradition: An Empirical Study of Sound Symbolism , 2003 .

[2] Fabien Gouyon,et al. Exploration of techniques for automatic labeling of audio drum tracks instruments , 2001 .

[3] Jian Tang,et al. Parametric vector quantization for coding percussive sounds in music , 2003, 2003 International Conference on Multimedia and Expo. ICME '03. Proceedings (Cat. No.03TH8698).