| 1. |
- Ferreol, A., et al.
(författare)
-
Modeling error sensitivity of the MUSIC algorithm conditioned on resolved sources
- 2008
-
Ingår i: European Signal Processing Conference. - 2219-5491.
-
Konferensbidrag (refereegranskat)abstract
- When the correlation matrix is known, the resolution power of subspace algorithms is infinite. In the presence of modelling errors, even if the correlation matrix is known, sources can no longer be resolved with certainty. Focusing on the MUSIC algorithm [1], the purpose of this work is to provide closed form expression of bias and variance versus the model mismatch (these errors can be different for each source). Un-like previous work, these performance measures are derived conditioned on the success of a certain source resolution test. Among the resolution definitions proposed in [2], we investigate which one is more suitable for our purposes. Numerical results support the theoretical investigations. Our findings are of a great interest for the determination of the necessary antenna calibration accuracy to achieve specifications on the estimator performance.
|
|
| 2. |
- Garcia, Guillermo, 1976, et al.
(författare)
-
A Methodology for speaker-dependent acoustic features based on a simplified cortical response for speaker verification systems
- 2009
-
Ingår i: 17th European Signal Processing Conference, EUSIPCO 2009; Glasgow; United Kingdom; 24 August 2009 through 28 August 2009. - 2219-5491. ; , s. 1240-1243
-
Konferensbidrag (refereegranskat)abstract
- Recently, the incorporation of the research done in biologically inspiredsystems has shown satisfactory results. A promising researcharea is the understanding of the human auditory systems and itsperformance under noisy conditions. Moreover, the incorporationof brain functions (cortical response) as an active part of the auditorysystem seems a viable alternative to increase the robustness ofspeech and speaker recognition systems. In this study, we proposea simplified model of the mammalian central auditory system forspeaker verification systems. This model is based on a dimension expansionrepresentation, attempting to capture the response of thecortical cells. Then, by means of the Principal Component Analysis(PCA) approach, we reduce the dimensionality and create speakerdependentfeatures. Our results showed that by using our modelingtechnique, we were able to improve the performance of speaker verificationsystems.
|
|
