This paper proposes the identification and separation method of multiple sound sources on a single moving object. The proposed method is based on the compensation of the Doppler effect for a specific sound source and uses the difference of temporal characteristics between the specific sound source and others. After deriving both of observed and compensated spectra in a general form, we discuss the frequency shift of two sound sources model. Then simple models composed of two moving sources which radiate sinusoidal waves are simulated, and the results show the effectiveness of the proposed method. The proposed method is effective only for narrow band or sinusoidal signal, because this method depends on the temporal characteristics of frequency shift due to the movement of sound sources.

View full abstract

For evaluation of aircraft fly-over noise, the yearly average noise level is required at many sites around an airfield. In the case of a military base, the operation of aircraft is usually irregular and the result of noise measurement during a short-period (one or two weeks) by itself cannot be a reliable estimate. This paper reports a procedure for estimation of the yearly average noise level at an arbitrary site around the airfield from the data of the short-term measurement by comparing with the observed results at a reference monitoring station. Most of airfields provide with monitoring installations near the runway ends and the aircraft fly-over noise is monitored all the year round. The measured noise level at an arbitrary site has a close relation to that of a nearby monitoring station, and the difference of short-term average noise levels seems to reflect that of longterm average noise levels between the two locations. The procedure for the estimation is similar to the determination of sound power levels of machinery noise sources using a reference sound source. It was examined by collected data obtained at many sites around several civil airports and military bases.

View full abstract

The effective reference pattern set structure and the training procedure for Japanese syllable recognition are reported. The studies of multi-reference pattern and training procedure were conducted for the optimum reference pattern structure to cover wide utterance variations. The examined training procedure included the sequential training to compensate the time dependent variation of pronounced pattern. The experiments proved that the recognition accuracy depends on the number of training rather than the number of reference pattern set, and also that the aggravation of recognition accuracy caused by the varied utterance with time is improved by the sequential training conducted after every recognition step.

View full abstract

As is well-known, the conventional standard identification methods for stochastic systems are usually employed under the optimum selection of the system order and system parameters, by use of the linear correlation technique and the usual least squares error method. In this paper, a new trial of finding some object-oriented system identification method matched to the prediction of output response probability distribution has been proposed in the functional and hierarchical forms. More explicitly, the indoor sound system on an energy scale is firstly described by a linear time series model supported by the well-known Sabine's equation or Statistical Energy Analysis method. Then, a generalized series expansion type probability density expression matched to an energy state variable has been proposed for the system output response in the form matched to the variety of fluctuation form, under the introduction of statistical information reflected hierarchically in expansion coefficients. Furthermore, instead of the usual system order and system parameters in the time series type system model, several kinds of functionally introduced system orders and system parameters have been estimated hierarchically in the matched form to the above hierarchical structure in the probability density expression. Finally, the validity and the effectiveness of the proposed theory have been experimentally confirmed too by applying it to the actual data observed in a reverberation room.

View full abstract

The transient response of our hearing system to the loudness of a rectangular pulse with a width of 0.05 ms was determined by the paired loudness comparison of two successive pulses with a varying time interval between them to that of a single pulse. The transient response was predicted to go a negative after an initial positive rise. The interference part of the transient consisting of a steep positive rise followed by a negative reaction was determined, and the loudness increase supposed to be due to the predominant pure tone spectrum caused by a time interval was also found. In the region from 50 to 80 ms of time interval, what was heard elsewhere as two separate tones was heard as three connected pulses.

View full abstract

It was reported that in the early stage of monaural hearing, the hearing system has an interfering transient response on loudness to the 0.05ms rectangular pulse excitation and there the loudness to each pulse is added having its sign. However, at binaural hearing, the loudness of two pulses of different signs from the opposite lateral directions was summed in the absolute manner and it was almost the same as the loudness of two rectangular pulses in the same sign or a single pulse of double the amplitude from one lateral direction. Loudness of reflected sound fields caused by two different sized plane panels having the direct pulsive sound were compared. The loudness difference between them was not explained by the difference of the largest instantaneous loudness, but was done by the integration of the loudness amplitude in the time window. In the last, the transient process on the hearing system for the loudness of 0.05 ms rectangular pulse was summarized with new understandings in the earlier and present paper.

View full abstract