Kensaku Fujii

A new noise reduction method using linear prediction error filter and adaptive digital filter

A technique that uses a linear prediction error filter (LPEF) and an adaptive digital filter (ADF) to achieve noise reduction in a speech degraded by additive background noise is proposed. Since a speech signal can be represented as the stationary signal over a short interval of time, most of speech signal can be predicted by the LPEF. On the other hand, when the input signal of the LPEF is a background noise, the prediction error signal becomes white. Assuming that the background noise is generated by exciting a linear system with a white noise, then we can reconstruct the background noise from the prediction error signal by estimating the transfer function of noise generation system. This estimation is performed by the ADF which is used as system identification. Noise reduction is achieved by subtracting the reconstructed noise from the speech degraded by additive background noise.

Method to update the coefficients of the secondary path filter under active noise control

In active noise control systems using a filtered-x algorithm, the secondary path must be identified before the coefficients of the noise control filter are updated. The secondary path, however, has an impulse response that is continually changing in practical systems. This change degrades noise reduction and the stability of the system. Therefore, the coefficients of the secondary path filter must be adjusted to actual impulse response samples at specific intervals. This paper proposes a method to update the coefficients under active noise control without feeding extra noise to the secondary source. Instead, this method uses estimation errors. The coefficients of the noise control filter generally have the different estimation errors whenever the coefficients are updated. To use the estimation error, this method implements the additional filter modeled on the overall path from a detection sensor to an error sensor, through the primary path, the control filter and the secondary path. Two different coefficient sets of the noise control filter derive two equations on this overall path. A solution of these concurrent equations naturally yields the impulse response samples of the secondary path. This paper uses a system identification technique to solve the concurrent equations. This technique is practical in that calculation errors are distributed equally over all coefficients of the secondary path filter. Finally, computer simulations explained in this paper confirm that the concurrent equation method can refresh the coefficients under active noise control.

Convergence time reduction provided by a block length control method applied to the “summational” NLMS algorithm

It is known in the normalized least mean square (NLMS) algorithm that the convergence time can be reduced by applying a control in which the step gain is gradually reduced, as the estimation precision is improved. A problem in realizing this idea is that the instantaneous value of the estimation precision which is needed in the control of the step gain cannot be observed directly from the residual signal which contains external disturbances. This paper considers a block length control, which can be used in the “summational” NLMS method. The method is able to improve the convergence of the estimation precision, as in step gain control, and a method of reducing the convergence time is proposed, utilizing the fact that saturation of the convergence behavior can be observed iteratively by extending the block. In addition, it is highly likely in an actual adaptive system that the reference signal, the power of the external disturbance, and the impulse response of the unknown system to be estimated as the coefficients of the finite impulse response (FIR) filter will change. It is not realistic to ignore the possibility of these changes in discussing the control of convergence. In the last part of this paper, it is shown that the proposed control method can handle these changes.

Multiple channel active noise control system based on simultaneous equations methods

For complicated noise fields, a multiple channel active noise control (ANC) system with several secondary sources, error sensors, and reference sensors is required. A multiple error filtered-X (MEFX) algorithm is developed for an adaptive algorithm in the multiple channel ANC system. This algorithm requires a previous estimation of the error paths. However, the characteristics of the error paths should be assumed to change during the operation of the system. This paper proposes a control algorithm for the multiple channel ANC system without the previous estimation of the error paths. This algorithm becomes an extension of the simultaneous equations method proposed for the single channel ANC system. In the proposed method, the good performances for noise cancellation and tracking of the changes in the error paths are achieved.

A hierarchical video compression method using object coding

In video communication applications, there are great demands of high-efficiently coding of video sequences. To improve subjective video quality, preferred ROT (Region of Interest) coding has widely been studied. In this paper, we employ the object coding techniques to encode ROT with higher priority. When we use the object coding to specify ROT, shape of the object does not require high accuracy. Therefore, we can improve the coding efficiency and reduce tracking difficulty by simplification of ROI shape.

Method to update the feedback control filter coefficients under active noise control

The active noise control system requires the process of identifying the secondary and the feedback paths previous to updating the coefficients of the noise control filter. These paths, however, are continuously changing in practical systems after the identification. This change inevitably degrades the performance of the system. The authors have already presented a method to update the coefficients of the secondary path filter modeling the former path under active noise control. This paper hence proposes another method to refresh the coefficients of the feedback control filter modeling the latter path. For this refreshment, the method forms two independent equations by using the estimation error involved in the coefficients of the noise control filter. This paper also introduces a system identification technique to solve the equations.

A new noise reduction method using linear predictor and adaptive filter

In this paper, we extend our previously proposed algorithm entitled Structural Maximum Likelihood Eigenspace Mapping (SMLEM) for rapid speaker adaptation. The SMLEM algorithm directly adapts Speaker Independent (SI) acoustic models to a test speaker by mapping the mixture Gaussian components from a SI eigenspace to Speaker Dependent (SD) eigenspaces in a maximum likelihood manner, with very limited adaptation data. In previous SMLEM paper, we presented encouraging results for SMLEM by adapting only the static feature components. In this paper, we propose a multi-stream approach where the static and dynamic feature streams are adapted. For small amounts of adaptation data ranging from 15 to 50 seconds, superior performance is demonstrated over both standard MLLR and block diagonal MLLR.A technique that uses a linear prediction error filter (LPEF) and an adaptive digital filter (ADF) to achieve noise reduction in a speech degraded by additive background noise is proposed. Since a voiced speech can be represented as the stationary periodic signal over a short interval of time, most of voiced speech is predicted by the LPEF. On the other hand, when the input signal of the LPEF is a background noise, the prediction error signal becomes white. Assuming that the background noise is represented as generate by exciting a linear system with a white noise, then we can reconstruct the background noise from the prediction error signal by estimating the transfer function of noise generation system. This estimation is performed by the ADF which is used as system identification. Noise reduction is achieved by subtracting the noise which is reconstructed by the ADF from the speech degraded by additive background noise.

Sub-RLS algorithm with an extremely simple update equation

A new type of adaptive algorithm is derived from a first order infinite impulse response (IIR) filter expression of the normalized least mean square (NLMS) algorithm. This new algorithm provides a convergence property similar to that of the recursive least square (RLS) algorithm. Its update equation, however, is extremely simple compared to that of the RLS algorithm. The new algorithm, named the sub-RLS algorithm, can be also derived from the least square (LS) algorithm on an approximation. The prefix sub designates the approximation applied to the LS algorithm for its recursive adaptation. This paper also presents a variation of reducing its processing cost.