Hazardous sounds represent a particular type of danger signal and should be heard clearly to ensure safety. Hearing-impaired people can be involved in accidents. This is exacerbated by the fact that they may have dificulty hearing hazardous sounds. Therefore, they spend time with a hearing-assistance dog in their daily lives. However, the number of hearing-assistance dogs has decreased recently, and they are less known in society and they may even be prevented from entering public facilities. Therefore, an animal external ear model-based mobile device that can detect hazardous sounds has been proposed. This device can alert the hearing-impaired person. In this paper, the animal external attachment with a humanand a cat-external ear model-verified for hazardous sound detection accuracy by microphone sensor is presented. Microphone sensors with and without a human- and a cat-external ear model are used. An ambulance siren and a bicycle bell are used as the hazardous sounds, and the spectral envelopes of each recorded sound are compared. The results show that the animal external ear model-based microphone sensor can more efectively detect the frequency features of the ambulance siren and the bicycle bell. A microphone sensor attached an external attachment with a human-external ear model can detect 2 ∼ 3 kHz frequency features with the highest sound pressure compared to the other microphones, regardless of the distance between the microphone sensor and the sound source. Also, a microphone sensor attached an external attachment with a cat-external ear model can detect wide range of frequencies with the highest sound pressure compared to the other microphone sensors, regardless of the distance between the microphone sensor and the sound source.
Various sounds are encountered in our daily lives. Among them, hazardous sounds pose possible
danger signals. Our safety might depend on being able to hear hazardous sounds such as an
emergency bell or the siren of an emergency vehicle. However, hearing-impaired people cannot
hear these hazardous sounds. In 2016, there were about 300,000 hearing-impaired people in
Japan [
Our new method can detect more hazardous sounds than only a microphone sensor, and alert
a hearing-impaired person to hazardous sounds, like the hearing-assistance dog. An animal
external ear model-based mobile device for hazardous sound detection has been proposed [
The rest of the paper is organized as follows. Section 2 outlines existing technologies. Section 3 explains the animal external ear model-based mobile device for hazardous sound detection. Section 4 describes how to process recorded sounds and its spectral envelope. Section 5 explains the experimental method. Section 6 shows the spectral envelopes of recorded sounds as experimental results, and Section 7 gives our conclusions.
There are four existing technologies to detect hazardous sounds. The first system detects
hazardous sounds and situations using clustering by the complete linkage method and probability
modeling of daily sounds [
The animal external ear model-based mobile device for hazardous sound detection is explained in this section. Overview of this device is shown in Fig. 1 (i). The hazardous sound is an alarm sound, siren, and emergency bell of an emergency vehicle which is designed to make people notice some danger. This device continuously pics up the sounds from the microphone sensor attached to an animal external attachment. The animal external attachment is shaped somewhat like the external ear of an animal (Fig. 1 (i) (a)). The picked-up sound is analyzed with a microcomputer and is determined to be hazardous sound (Fig. 1 (i) (b)). If the sound is judged to be a hazardous sound, the device notifies the hearing-impaired person using LED or other methods (Fig. 1 (i) (c)). This device is implemented as a mobile device.
Figure 1 (ii) shows the flow of recorded sounds processing in this experiment. The purpose
of this process is to derive the spectral envelope. A sound is recorded by the microphone
sensor. And, the recorded data are processed by a short-time Fourier transform (STFT) to find
a spectrogram [
CEP(k) = DFT(log|X(k)|2) From the cepstrum, the quefrency above a certain threshold is set to 0 to extract the low quefrency components of the cepstrum. Quefrency is a twisted phrase of “frequency” and horizontal axis of cepstrum. Such quefrency includes components of the spectral envelope, which is defined as log(|highpass(k)|2). Then, an inverse discrete Fourier transform (IDFT) is applied to the resulting data to find the spectral envelope, which configured as follows SPE(k):
SPE(k) = IDFT(log|highpass(k)|2)
The spectral envelope includes formants, which are frequency characteristics of a specific
frequency region that characterize the sound. In particular, the spectral envelope is useful
when indicating vocal-tract characteristics of a human voice. By observing formants can get
the frequency characteristics of a vocal tract [
intervals of 0.65 s. The bicycle bell is rung by hand. The sound sources are recorded for 15 s,
and the recorded sounds are processed in section 3. Three microphone sensors are connected to
a SPRESENSE [
Figure 3 shows the spectral envelope of the ambulance siren for distances of 1 m and 3 m, respectively. The horizontal axis is frequency and the vertical axis is power. The power is decreased by about 3 dB due to change the distance between the microphone sensors and the sound source from 1 m to 3 m in Fig. 3. It is shown that each formant of the ambulance siren can be detected at around 0.77 kHz and 0.96 kHz at intervals of 0.65 s regardless of the distance between the microphone sensors and the sound source from Fig. 3. Also, Fig. 3 shows that MIC_human detects formant of 2 ∼ 3 kHz frequency features with the highest sound pressure compared to the other microphone sensors, regardless of the distance between the microphone sensor and the sound source. On the other hand, MIC_cat detects formants of wide range of frequencies with the highest sound pressure in the range of 0 ∼ 7.5 kHz compared to the other microphone sensors, regardless of the distance between the microphone sensor and the sound source. And the sound pressure detected by MIC_human and MIC_cat are almost greater than the sound pressure detected by MIC_only.
The spectral envelope of the bicycle bell is shown in Fig. 4 for distances of 1 m and 3 m, respectively. The horizontal axis is frequency and the vertical axis is power. The power is decreased by about 1 dB due to change the distance between the microphone sensors and the sound source from 1 m to 3 m in Fig. 4. Figure 4 shows that each formant of the bicycle bell can be detected five places at around 2.0 kHz, 3.6 kHz, 5.0 kHz, 8.5 kHz, and 15 kHz, regardless of the distance between the microphone sensors and the sound source. Also, Fig. 4 shows that MIC_human detects formant of 3.6 kHz with the highest sound pressure compared to the other microphones regardless of the distance between the microphone sensor and the sound source. On the other hand, MIC_cat detects formants of 5.0 kHz, 8.5 kHz, and 15 kHz with the highest sound pressure compared to the other microphones, regardless of the distance between the microphone sensor and the sound source. In other words, MIC_human can detect approximately 2 ∼ 3 kHz frequency features and MIC_cat can detect a wide range of frequency features in the range of 0 ∼ 18 kHz from Fig. 4. And, the sound pressure detected by MIC_human and MIC_cat are almost greater than the sound pressure detected by MIC_only.
Our proposed method can detect more hazardous sounds more than only a microphone sensor, and alert a hearing-impaired person to hazardous sounds, like the hearing-assistance dog. In this paper, the animal external attachment with a human- and a cat-external ear model-verified hazardous sounds detection accuracy by the microphone sensor. Three diferent microphone sensors are compared, and the spectral envelopes of two recorded sounds are detected. The results show a microphone sensor attached an external attachment with a human-external ear model can detect 2 ∼ 3 kHz frequency features with the highest sound pressure compared to the other microphone sensors, regardless of the distance between the microphone sensor and the sound source. Also, a microphone sensor attached an external attachment with a cat-external ear model can detect wide range of frequencies. This enables the highest sound pressure possible compared to the other microphone sensors, regardless of the distance between the microphone sensor and the sound source. Thus, the animal external ear model-based microphone sensor can more efectively detect the frequency features of the ambulance siren and the bicycle bell. Therefore, it makes easier to detect hazardous sounds by adapting the external attachment of the microphone sensor to match each hazardous sound in our proposed device. In the future, experiments will be conducted in various environments, including noisy environments and changing temperatures. In addition, various external ear models will be verified to detect hazardous sounds more efectively by changing the shape and material of external attachment. Furthermore, efectiveness of our proposed method which detects other hazardous sounds that are common in daily life is verified.