This paper describes a method of estimating sound source contribution factors; it utilizes envelopes of band-passed signals picked up at the noise source and at the observation point. The coherence function has been used to express the contribution factor of a source signal to an observation point signal. In the open air, fluctuation of the transmission time caused by wind and/or air turbulence decreases the value of the coherence function; in other words the estimated transfer characteristics contain considerable error. The effect of such fluctuations increases in proportion to the length of transmission path and the frequency of signals. Thus, the authors propose in this paper the use of envelopes of band-passed signals instead of the original wave forms for estimating sound source contribution factors. With the cross spectral technique applied to those envelopes, the power transmission rate and the contribution factor of each frequency band can be obtained without being affected by a fluctuation of the transmission time. The method proposed here is useful for noise source detection.

View full abstract

A practical method for the qualitative analysis of impaired speech perception was developed based on graphical representations of psychoacoustic proximity among monosyllables. The psychoacoustic proximity was defined as the degree of ease of confusion among monosyllables for normal ears, and was measured as a confusion matrix among 67 Japanese monosyllables called “chokuon.” This matrix was constructed from hearing tests on unclear perception induced by lowering stimulus levels for 26 listeners with normal hearing. Configuration graphs of the monosyllables where the proximity values were represented by the mutual distances were made from the confusion matrix using a non-metric multi-dimensional scaling method known as SMACOF. Using these graphs, the perceptual failures for the monosyllables on the part of a patient can be simply documented by connecting confused stimuli and corresponding wrong responses. This documentation indicates the abnormality of a patient's failures in terms of discrepancies from the ease of confusion for normal ears. Using two hearing-impaired patients, it was concluded that the analysis of these discrepancies together with the patients' pure tone audiogram provides useful information to know how and why patients fail to perceive certain speech sounds.

View full abstract

A Japanese text input system by speech that is uttered character by character is described. The system mainly consists of an acoustic processing stage and a linguistic processing stage. Acoustic processing is carried out through speech analysis based on linear prediction, segmentation of input speech into syllable strings and syllable recognition based on a dynamic programming method. Translation of syllable recognition results into Japanese text is done by the processes of word candidate retrieval and word ranking through the use of a Kanji-Kana word dictionary and grammatical rules. To cover for segmentation errors which occur due to inadequate handling in coarticulation, two kinds of error correction algorithms are adopted: One being syllable boundary correction by phonemic distortion rules in the acoustic processing stage and the other being recognition result revision in the linguistic processing stage. The system was tested by running it through acoustic and linguistic processing experiments using a Japanese text uttered by five speakers. A syllable recognition rate of 86.7 % and a Japanese translation rate of 83.1 % were obtained.

View full abstract

This paper describes a new automatic method for detecting defects in ball bearings. The detection of defects is currently carried out by inspectors who listen to vibration signals obtained by a vibration pick-up in the Anderon meter. The pick-up is attached to the outer ring of a ball bearing while the inner ring rotates at a uniform speed. Several methods have been proposed so far for the automatic detection of flaws. These methods are based on the periodicity of vibration pulses excited by flaws. However, the periodicity of vibrations is not always guaranteed when (a) there are slight flaws on the surface of the race, (b) there are flaws on the surface of balls, and (c) there is dust in the grease. However, we have developed a new method, by which the non-periodic resonant vibrations due to both slight flaws and dust are detected. We have applied this new method to the detection of these defects in small-sized ball bearings; the defects were detected with an 98 % accuracy rate.

View full abstract