Hazardous sounds exist in our daily lives. There is a possibility of encountering accidents if people cannot hear hazardous sounds. However, hearing-impaired people cannot hear these sounds and may be at risk of accidents. Therefore, some hearing-impaired people live with a hearing-assistance dog. However, the number of hearing-assistance dogs has decreased recently. Therefore, animal ear-like hazardous sound detection devices have been proposed to support hearing-impaired people. Relevant hazardous sounds include sudden sounds and approaching sounds. If these sounds can be detected and communicated to hearing-impaired people, hearing-impaired people can be aided in noticing danger around them and take action to remain safe. Furthermore, such a device can help alert hearing-impaired people about danger before it happens and to avoid it. Additionally, when the device detects hazardous sounds, it is necessary to distinguish them from surrounding noises. This paper proposes a method to distinguish the hazardous sounds using the Zero Crossing Rate (ZCR), to be implemented for recognition on the proposed device. In the experiment, a variety of sounds are recorded and categorized as sudden sounds, approaching sounds, and surrounding noises. The ZCR of each sound is calculated from the recorded sound data. The results show that the ZCR waveform can be said that it reflects the characteristics of the waveform of the sound. It is demonstrated that there are diferences in the ZCR values between the sudden sounds, the approaching sounds, and the surrounding noises. Therefore, the ZCR may be a useful aid in distinguishing hazardous sounds. The ZCR can distinguish hazardous sounds regardless of their sound pressure. Also, the advantage of using ZCR is that it is independent of any specific frequency threshold when detecting. Furthermore, the ZCR is a feature that requires little computation because it is only based on counting how many times the signal changes sign. For this reason, it is suitable for real-time processing on microcomputers.
Hazardous sounds exist in our daily lives. There is a possibility of encountering accidents if
people cannot hear hazardous sounds. However, hearing-impaired people cannot hear these
sounds. In 2022, there were about 310,000 hearing-impaired people in Japan [
Relevant hazardous sounds include sudden sounds and approaching sounds. Example of sudden sounds include alarms, bicycle bells, something falling down, and so on. These sounds are signs of danger and environmental change. On the other hand, approaching sounds are sounds made by approaching trains, bicycles, cars, and so on. These sounds also indicate potential danger. They both help us notice danger around us quickly. So, it is important to pay attention to these sounds and take action to remain safe in daily life. Surrounding noises (aka, ambient noise) are the background sounds that you hear in everyday places. This noise is always there, even if you don’t notice it. Surrounding noises can make it dificult to hear the sounds you need to be aware of. If these various sounds can be detected and communicated to hearing-impaired people, hearing-impaired people can be aided in noticing danger around them and take action to remain safe. Furthermore, it can help hearing-impaired people know about danger before it happens and avoid it. Also, when the device detects hazardous sounds, it is necessary to distinguish them from surrounding noises. This paper proposes a method to distinguish the hazardous sounds to be implemented for recognition on the proposed device. This study specifically investigates a distinguishing approach based on the Zero Crossing Rate (ZCR).
The rest of the paper is organized as follows. Section 2 explains existing technologies. Section 3 describes the animal ear-like hazardous sound detection device. Section 4 outlines the Zero Crossing Rate (ZCR). Section 5 explains experimental method consists of the prototype of animal ear-like hazardous sound detection device, recording environment, and calculation of the ZCR. Section 6 shows the ZCR values as experimental results, and Section 7 gives our conclusions.
There are four existing technologies that are related to either detecting hazardous sounds or
supporting hearing-impaired people. The first system identifies hazardous sounds and situations
through clustering based on the complete linkage method and probabilistic modeling of daily
environmental sounds [
The animal ear-like hazardous sound detection device is explained in this section. An overview of this device is shown in Fig. 1. Hazardous sounds assist people to notice danger. The processing lfow of the proposed device is shown below.
1. The device continuously gets sound from the microphone sensor attached an animal external attachment (Fig. 1 (i)). 2. The acquired sound is analyzed with a built-in microcomputer and is judged to be a hazardous sound or not (Fig. 1 (ii)). 3. If the sound is judged to be a hazardous sound, the device notifies the hearing-impaired people using methods such as LEDs and vibrations (Fig. 1 (iii)).
The animal external attachment is shaped somewhat like the external ear of an animal.
Because animals have two ears, they are able to judge the direction and source of sounds.
Animals can detect enemies and danger in real time by using their ears. Their hearing abilities
vary widely across species. That said, animals have evolved their ears for sound localization,
which helps them detect and respond quickly to enemies and danger [
The zero crossing is the number of times the sound waveform crosses the zero level. Specifically,
the Zero Crossing Rate (ZCR) is defined as the proportion of zero crossings within a single
frame like Fig. 2. The ZCR is used to distinguish voiced and unvoiced signals [
ZCR()[%] = 100 ∑−1︁ 1 =1 2 |sgn([ + ]) − sgn([ − 1 + ])| (1) The frame length is samples, and the frame shift length is samples. The frame shift is the number of samples between frame starts. The function sgn() in equation. (1) is defined equation. (2).
⎧1 ⎪ sgn() = ⎨
> 0 0 = 0 ⎪⎩−1 < 0 (2)
|sgn([ + ]) − sgn([ − 1 + ])| This term takes the value 1 if the signs of two consecutive samples difer, and 0 otherwise. The factor 12 is used to make the term equal 1 when a zero crossing occurs, because the diference of the sign values becomes ± 2 in that case. Adding this term from = 1, . . . , − 1 gives the total number of zero crossings in the frame. Dividing this number by normalizes it according to the frame length. The result of equation. (1) is multiplied by 100 to express the ZCR as a percentage.
Consequently, the ZCR provides a simple and efective measure of the spectral characteristics of the signal. It is widely used in speech processing tasks such as voiced and unvoiced classification and noise estimation. In addition, the ZCR is considered to be efective for distinguishing waveform 0
Frame1
Frame shift
Frame2
such as the sudden sounds, the approaching sounds, and the surrounding noises. Its applicability to these sounds is examined in this study.
5.1. The prototype of animal ear-like hazardous sound detection device
A prototype of the animal ear-like hazardous sound detection device is used in the experiment.
An overview of the prototype device’s external appearance and inner structure is shown in
Fig. 3 (a). The prototype device includes a SPRESENSE microcomputer, consisting of a main
board (CXD5602PWBMAIN1) and an extension board (CXD5602PWBEXT1) [
1 2 3 Time4[s] 5 6 7 1 2 3 Time4[s] 5 6 7 8 []ZCR%1111(648224006b0) ZCR1wavef2oMromtorc3ycTliinmgeh4[osr]n 5 Mo6torcycl7ing hor8n []ZCR%1111(648224006b0) ZACpRp1rwoaacvheifn2ogrmmoto3rcTyicmlee4[s] Mov5ing aw6ay mot7orcycle8 (i) The sudden sounds (motorcycling horn sound) (ii) The approaching sounds (approaching motorcycle sound) sound and a rapid decrease when the motorcycling horn sound ended. In Fig. 4 (ii) (b), the ZCR waveform of the approaching sounds shows a gradual increase as the motorcycle approaches and a gradual decrease as it moves away. This can be considered a result of the Doppler efect. In Fig. 4 (iii) (b), the ZCR waveform of the surrounding noises shows small fluctuations, but an increasing or decreasing trend is not observed. These results show that the ZCR waveform reflects the characteristics of the waveform of the sound itself. Also, Fig. 4 (i) and (ii) show that an increase of sound amplitude does not lead to an increase of the ZCR.
Table. 1 shows the average of the ZCR from particular sound sections of 10 samples for each type of sound, including sudden sounds, approaching sounds, and surrounding noises. The average of the ZCR for the sudden sounds is 14.40 %. The average of the ZCR for the approaching sounds is 6.70 %. The average of the ZCR for the surrounding noises is 1.48 %. These results show that there are diferences in the ZCR values between the sudden sounds, the approaching sounds, and the surrounding noises. Furthermore, it is confirmed that the ZCR can be obtained in noisy environments without being afected by noise. Therefore, the ZCR may be a useful aid in distinguishing hazardous sounds. The ZCR can distinguish hazardous sounds regardless of their sound pressure. Also, hazardous sounds vary greatly, making it dificult to limit detection to specific frequencies. However, the ZCR has the advantage of being independent of any specific frequency threshold when detecting. Furthermore, the method to distinguish the hazardous sounds using the ZCR is considered applicable without the need for additional filtering processes. The ZCR is a feature that requires little computation because it is only based on counting how many times the signal changes sign. For this reason, it is suitable for real-time processing on microcomputers.
This paper proposes a method to distinguish the hazardous sounds using the Zero Crossing Rate (ZCR), to be implemented for recognition on the proposed device. For the experiment, three types of sounds were investigated:1) sudden sounds; 2) approaching sounds; and 3) surrounding noises are recorded using the prototype of an animal ear-like hazardous sound detection device. Subsequently, the ZCR of each sound is calculated from the recorded sound data. The results show that the ZCR waveform reflects the characteristics of the waveform of the sound itself. Furthermore, there are diferences in the ZCR values between the sudden sounds, the approaching sounds, and the surrounding noises. Therefore, the ZCR may be a useful aid in distinguishing hazardous sounds. The ZCR can distinguish hazardous sounds regardless of their sound pressure. Also, the advantage of using ZCR is that it is independent of any specific frequency threshold when detecting. Furthermore, the ZCR is a feature that requires little computation because it is only based on counting how many times the signal changes sign. For this reason, it is suitable for real-time processing on microcomputers.
In the future, sounds that are not used in the experiment will also be tested. For example, emergency sirens and falling objects. In addition, recording is conducted in environments with rain and wind, as well as in situations where many sounds overlap. It is verified whether detection correctly under these conditions. Furthermore, this method is planned to be implemented in animal ear-like hazardous sound detection devices. Finally, the goal is to develop a device that can more accurately detect hazardous sound and more broadly assist the hearing-impaired people in real-world conditions.