7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Time-Frequency Transforms and Beamforming for Speaker Recognition

Antonio Satué-Villar (1), Juan Fernández-Rubio (2)

(1) Escuela Universitaria Politécnica de Mataró, Spain; (2) Universidad Politécnica de Cataluña, Spain

In this paper we study the advantages of a data transformation using time-frequency transforms (wavelets, in our case) in a wideband beamforming system. For this purpose, we simulate a target signal and interfering jammers that are impinging upon a microphone array in a noisy environment. The target and interfering signals are obtained from a speech database. The simulations presented show that the wavelet transform reduces the time of convergence, even though there is not an unique optimal wavelet filter; it depends on the speech signal. In this paper it can be also infered that the wavelet transform do not affect substantially to the quality of the output signal (evaluated with a speaker recognition method).


Full Paper

Bibliographic reference.  Satué-Villar, Antonio / Fernández-Rubio, Juan (2002): "Time-frequency transforms and beamforming for speaker recognition", In ICSLP-2002, 2533-2536.