This report describes the basic study of sound field control method using multi-channel loudspeakers. The study is motivated by the needs for constructing the trans-aural system in reflective or reverberant environments. We utilized least squared error approach and calculated inverse filter coefficients of sound fields. Our simulation results suggest the conjugate gradient method is suitable for the sound fields control by multi-channel system which requires the long FIR filter taps.

View full abstract

Window-based measurement and simulation system was developed. The system supports the facilities such as (1) measurements of acoustic system response, (2) wave editor for the measured responses, (3) system analyses, (4) sound field simulation, and (5) GUI-based software environment for visualization and presentation. These functions provide powerful tools to design, modify and auralize the sound fields such as concert hails and recording studios. The system is expandable for image processing as well as acoustic signal processing. It is therefore possible to construct the generalized system for sound and image processing that will be required in the future information processing.

View full abstract

This paper describes the effects of time windows on the impulse response estimation by the cross spectral method. The effect of window shape is first analyzed and the effect of the finite window length is next analyzed. The amplitude of the impulse response estimated by the cross spectral method is decreased at the central part of the time window by the effect of window shape, and the amplitude is descreased by the effect of the finite window length with the laps of time. The results of the analyses are confirmed by the computer simulation using random sequence as the source signal. According to the investigation, it is made clear that the highest estimation accuracy is obtained when the rectangular window is used and the accuracy is much improved by correcting the estimated impulse response using correction factor given in this paper especially in the case of short time window.

View full abstract

This paper proposes a new method for the estimation of the magnitude of sinusoidal components by FFT analysis. The magnitude of a sinusoidal wave estimated by FFT takes a value less than the actual value if the number of waves in the time window is not integer. A new method is proposed to obtain accurate magnitude. The effectiveness of the proposed method is illustrated by the computer simulation. Finally, the problem in the power spectrum estimation using the proposed method is discussed.

View full abstract

previously, we proposed a remote control system for television set as an example of an isolated word recognition system for noisy environment. In this paper, we discuss a speech parameter extraction method using Masking model in order to improve the performance of the proposed system. As well known, Masking effect is one of basic phenomena on human auditory system and it can be treated as a noise suppression mechanism for speech recognition process of human. The proposed system can recognize over 90% of words under -15dB of SNR.

View full abstract

An accurate estimation of an environment noise level is essential to an accurate sound level control in an audio system that compensates the loss of the loudness of reproduced Sound caused by the masking of the noise. We studied a noise level estimation method using microphone with rejecting the reproduced sound element from the microphone signal. We made up a hardware for examination using a digital signal processor which works as two channel adaptive filters cope with two channel stereo signals. The adaptive filters were composed of two 64 stage FIR type filters,and the data rate was selected to 2.44 ksps. As a result,20dB rejection of the sound signal has been obtained which satisfies the requirement from an automatic sound level control in an audio system.

View full abstract

This paper presents a real time pitch detector for speech, especially singing voice. In this pitch detector, the fundamental period is determined by zero-crossing interval of speech signal regardless of having zero-crossing point caused by 2nd or 3rd harmonics. The pitch detector consist of a fixed-point digital signal processor and an 8-bit microcomputer. According to our experiment, 95.4% of mean pitch detection rate was obtained for vowels.

View full abstract

In conventional theories of Active Noise Control (ANC), there is a limit to explain ANC, especially in three-dimensional sound field that is recently tried experimentally. We clarify the reason of the limitation by comparative studying of those conventional ANC theories. In order to express ANC theory that we have already studied from energetic viewpoint, we use the Kirchhoff's integral equation.

View full abstract