Il Whan Cha

A new voice transformation method based on both linear and nonlinear prediction analysis

We describe a voice transformation method which changes the source speakers acoustic features to those of a target speaker. In the method acoustic features are divided into two parts, linear and nonlinear parts. Linear parts are characterized by LPC cepstrum coefficients which are obtained from LP analysis. The nonlinear part, which represents the excitation signal, is modelled by the long-delay nonlinear predictor using a neural net. Conversion rules for the excitation signal are generated by the average pitch ratio and the mapping codebook, and those for LPC cepstrum coefficients are based on the orthogonal vector space conversion. In addition, the spectral envelope compensation is proposed to correct spectral distortion. In the transformed speech a listening test shows that the proposed method makes it possible to convert speakers individuality while maintaining high quality.

An active noise control algorithm for controlling multiple sinusoids

The filtered-x LMS algorithm and its modified versions have been successfully applied in suppressing acoustic noise such as single and multiple tones and broadband random noise. This paper presents an adaptive algorithm based on the filtered-x LMS algorithm which may be applied in attenuating tonal acoustic noise. In the proposed method, the weights of the adaptive filter and estimation of the phase shift due to the acoustic path from a loudspeaker to a microphone are computed simultaneously for optimal control. The algorithm possesses advantages over other filtered-x LMS approaches in three aspects: (1) each frequency component is processed separately using an adaptive filter with two coefficients, (2) the convergence parameter for each sinusoid can be selected independently, and (3) the computational load can be reduced by eliminating the convolution process required to obtain the filtered reference signal. Simulation results for a single-input/single-output (SISO) environment demonstrate that the proposed method is robust to the changes of the acoustic path between the actuator and the microphone and outperforms the filtered-x LMS algorithm in simplicity and convergence speed.

Realization of a generalized sidelobe canceller

The use of adaptive multichannel lattice/escalator filters to realize a generalized sidelobe canceller is proposed. This paper studies the performances of the proposed methods and the TDL counterpart via computer simulations for both stationary and nonstationary environments. Experimental results show that the lattice/escalator realizations converge faster and yield smaller mean-square errors than the TDL counterpart. Also, experiments seem to indicate that the beam patterns of the escalator realization are the most consistent, while those of the lattice realization are more consistent than the TDL counterpart.

On the performance of the generalized sidelobe canceller in coherent situations

The convergence rate for the adaptive weights to reach the optimum value in an adaptive array system depends on the eigenvalue spread ratio of the autocovariance matrix. How the eigenvalue spread ratio in a generalized sidelobe canceller (GSC) is affected by the various parameters is studied. Expressions for the output power of the GSC in coherent situations are derived, and numerical results are included. >

The output SINR of the linearly constrained beamformer

In this paper, expressions for the output signal-to-interference plus noise ratio(SINR) of the linearly constrained narrowband beamformer in noncoherent sityations are derived using a verctor approach.