Solving global permutation ambiguity of time domain BSS using speaker specific features of speech signals
| Autor: | Ali Khanagha, Vahid Khanagha |
|---|---|
| Přispěvatelé: | Geometry and Statistics in acquisition data (GeoStat), Inria Bordeaux - Sud-Ouest, Institut National de Recherche en Informatique et en Automatique (Inria)-Institut National de Recherche en Informatique et en Automatique (Inria), Yahia, H. |
| Jazyk: | angličtina |
| Rok vydání: | 2009 |
| Předmět: |
Computational complexity theory
business.industry Speech recognition Feature extraction SIGNAL (programming language) Estimator Pattern recognition Blind Source Separation(BSS) Multilateration Speech processing [INFO.INFO-SD] Computer Science [cs]/Sound [cs.SD] Multiple Speaker Localization [INFO.INFO-TS]Computer Science [cs]/Signal and Image Processing Mel-frequency cepstrum Artificial intelligence Time domain Particle Swarm Optimization(PSO) business [SPI.SIGNAL]Engineering Sciences [physics]/Signal and Image processing Mathematics |
| Zdroj: | 2009 IEEE Symposium on Industrial Electronics and Applications. 2009 IEEE Symposium on Industrial Electronics and Applications., Oct 2009, Kuala Lumpur, Malaysia |
| Popis: | International audience; Multidimensional localization of competing speakers using BSS based TDOA estimations, requires the solution of global permutation ambiguity before fusing several TDOA estimates. Since the separation quality of BSS is not perfect, it is not easy to decide which TDOA belongs to which source (specially when the number of speakers grows). We study the robustness of several speaker specific features of speech against dereverberation filtering, by evaluating their capability to recognize perceptually dominant sources in each one of moderately enhanced outputs of the BSS algorithm. We compare the performance of several features in terms of Average Decision Statistic and computational complexity. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|