Approaching Bicycle Sound Detection Using Weighted-Zero Crossing Rate Method for Animal Ear-like Hazardous-Sound Detection Device

Many hearing-impaired people are at risk of accidents because they cannot perceive hazardous sounds. Therefore, some hearing-impaired people rely on hearing assistance dogs. However, the number of such dogs is insufficient to meet demand, highlighting the need for alternative assistive technologies. Hazardous sounds in daily life can be classified into sudden sounds and approaching sounds, and it is important to help hearing-impaired people to distinguish between these types of sound so that they can take appropriate evasive action if necessary. Given the potential hazard posed by bicycles to hearing-impaired pedestrians, a method for detecting approaching bicycle sounds is therefore needed. Estimating the cyclingspeed of an approaching bicycle can permit hearing-impaired people to assess the hazard more precisely and receive appropriate alerts. An animal ear-like hazardous-sound detection device has been proposed to support hearing-impaired people. The device detects hazardous sounds and alerts the user. This paper verifies the characteristics of approaching bicycle sounds at different cycling-speeds using a weighted-zero crossing rate (weighted-ZCR) method with a prototype device. In the experiment, approaching bicycle sounds are recorded outdoors, and weighted-ZCR values are calculated from the recorded sounds. The results show that the weighted-ZCR values gradually increases as a bicycle approached and decreased as it moves away. Furthermore, the weighted-ZCR values varies with the cycling-speed, and the variance of the weighted-ZCR values changes with cycling-speed. It is well-known that the risk of injury increases sharply as the cycling-speed increases from 10 km/h to 15 km/h. Therefore, cycling-speeds of 15 km/h or higher are considered hazardous. Based on the experimental results, the prototype device can classify between cycling-speeds of 15 km/h or higher from those below 15 km/h using a receiver operating characteristic (ROC) curve based on the weighted-ZCR threshold. Classification performance is maximized at the weighted-ZCR threshold of 0.429 over distance of 1 - 3 m for both the right and left animal external ear model-based microphone sensors of the animal ear-like hazardous-sound detection device. The area under the ROC curve (AUC) is 0.985. These results indicate that the animal ear-like hazardous-sound detection device can detect approaching bicycles traveling at 15 km/h or faster within 3 m of hearing-impaired people.