Name
Abstract | ||
|---|---|---|
In extending previous work on detecting transient spectral peaks we here investigate into the distinction between sinu- soidal and noise components by means of classification of spectral peaks. The classification is based on descriptors de- rived from properties related to time-frequency distributions. In contrast to existing methods, the descriptors are designed to properly deal with non-stationary sinusoids, which consid- erably increases the range of applications. The experimental investigation shows superior classification results compared to the standard correlation-based approach. probabilities depend on the size of the analysis window (for a larger window more noise peaks will be observed in contrast to a constant number of of sinusoidal peaks) it appears more appropriate to adjust classification thresholds such that class dependent error rates are achieved. Consequently, we derive our classification criteria by means of declaring a worst case signal and limit the error rate for each of the different classes. There exist a number of audio signal processing applica- tions that may benefit from a correct classification of spec- tral peaks. Musically interesting is the possibility to sepa- rate sinusoidal and noise components by means of grouping the classified spectral peaks. As a further example we men- tion the possibility to reduce the number of candidate peaks considered for partial tracking which could reduce the com- putational costs for probabilistic partial tracking algorithms (Depalle, Garcia, and Rodet 1993). The paper is organized as follows. In section 2 we define the descriptors that will be used for classification of spectral peaks and discuss their properties if applied to different types of spectral peaks. In section 3 we describe the structure of the decision tree and derive the thresholds to be used for classi- fication. An experimental demonstration of the superior per- formance of the new classifier compared to the correlation based peak classification is presented. We conclude the paper with a discussion of the achievements in section 4. |
Year | Venue | Field |
|---|---|---|
2004 | ICMC | Pattern recognition,Correlation,Artificial intelligence,Mathematics,Decomposition |
DocType | Citations | PageRank |
Conference | 5 | 1.13 |
References | Authors | |
2 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Axel Roebel | 1 | 118 | 16.84 |
| Zivanovic, M. | 2 | 59 | 10.80 |
| Xavier Rodet | 3 | 627 | 107.87 |