抄録
In recent years, over one million loudspeakers are manufactured a month for TV sets, radio sets, etc.. The purpose of the loudspeaker is to produce high-fidelity sound. As abnormal sound is not permissible, all of them must be inspected at the end of production line. It is painful for human inspectors to listen to the fullpower sound.
In this paper, a method to inspect the loudspeaker by acoustic signal processing techniques. The spectral map shows that it is not necessary to sweep the frequency of the driving signal but sufficient to fix it to the free air resonance frequency (f0). In this system, f0 is measured by a loudspeaker tester and a sine wave of frequency f0 is generated by a micro-computer and a DA converter. Every loudspeaker is driven by this sinusoidal wave of its full power. Its sound, acquired synchronously to the driving frequency, is emphasized with a high frequency band-pass filter and digitized into 12 bits. Its power spectral distribution, obtained by Fast Fourier Transform through an array processor, is inspected by a linear classification method. In the classification process, we use the distance from the average power spectrum of normal loudspeakers. The number of abnormal loudspeakers tested is 38, which has been extracted during two days in a production line.
The result showed that the learning rate was 100% and the recognition rate was 89.5%. Misrecognized sounds were so critical that some human inspectors gave an undecisive classification. The automatic loudspeaker inspection method provides a better environment for human inspectors and is effective to quality control of the products.
