A Study on the Sound Quality Evaluation Model of Mechanical Air-Cleaners

Abstract

In operating the air-cleaner for a long time, people in a quiet enclosed space expect low sound at low operational levels for a routine cleaning of air. However, in the condition of high operational levels of the cleaner, a powerful yet nonannoying sound is desired, which is connected to a feeling of an immediate cleaning of pollutants. In this context, it is important to evaluate and design the air-cleaner noise to satisfy such contradictory expectations from the customers. In this study, a model for evaluating the sound quality of air-cleaners of mechanical type was developed based on objective and subjective analyses. Sound signals from various air-cleaners were recorded and they were edited by increasing or decreasing the loudness at three wide specific-loudness bands: 20–400 Hz (0–3.8 barks), 400–1250 Hz (3.8–10 barks), and 1.25–12.5 kHz bands (10–22.8 barks). Subjective tests using the edited sounds were conducted by the semantic differential method (SDM) and the method of successive intervals (MSI). SDM tests for seven adjective pairs were conducted to find the relation between subjective feeling and frequency bands. Two major feelings, performance and annoyance, were factored out from the principal component analysis. We found that the performance feeling was related to both low and high frequency bands, whereas the annoyance feeling was related to high frequency bands. MSI tests using the seven scales were conducted to derive the sound quality index to express the severity of each perceptive descriptor. Annoyance and performance indices of air-cleaners were modeled from the subjective responses of the juries and the measured sound quality metrics: loudness, sharpness, roughness, and fluctuation strength. The multiple regression method was employed to generate sound quality evaluation models. Using the developed indices, sound quality of the measured data was evaluated and compared with the subjective data. The difference between predicted and tested scores was less than 0.5 points.