Note:
This bibliographic page is archived and will no longer be updated.
For an up-to-date list of publications from the Music Technology Group see the
Publications list
.
Detection and Modeling of transient regions in musical signals
| Title | Detection and Modeling of transient regions in musical signals |
| Publication Type | Master Thesis |
| Year of Publication | 1999 |
| Authors | Gouyon, F. |
| preprint/postprint document | files/publications/DEA1999-gouyon.pdf |
| Abstract | It is well-known that regions corresponding to onsets and decays of notes are essential in the perception of a sound in order to determine its source recognition and its natural or synthetic character. Some important work has been and is being done regarding the detection of onsets in audio and the segmentation of signals, as well as regarding the characterization of musical events according to meaningful and practical features. Although many methods for signals representation exist, most methods do not represent certain problem regions coherently (attacks or decays of notes for example, but also other regions), and therefore can neither modify nor reproduce these regions in a meaningful way. Any accurate sound content description and analysis/synthesis application (pitch and time scaling, compression, hybridization...) calls for a unification of the characterization and segmentation issues. Thus, in this work, we consider parallel and interacting schemes for detection and modeling of nonstationarities, or transient regions in general, which arise from the polyphonic nature of the signal as well as from each instrument's separate performance. We present an analysis/synthesis framework involving an explicit parametric model for transients based on Prony's technique. Explicit modeling of transients should be deeply embedded with an accurate segmentation of the signal. Segmentation, here, is based on the detection of abrupt changes in signal statistics; this generalizes well-known techniques based solely on the detection of energy-jumps as the discriminant statistical parameter. |
