speaker performance on sports performance

-Behavior analysis using OpenPose- Satoshi Miharu 1 Kanagawa Institute of Technology 1030 Shimo-ogino, Atsugi City, Kanagawa, 243-0292 Japan Ryota Matsuishi 2 TOA Corporation 7-2-1 Minatojima-nakamachi, Chuo-ku, Kobe, Hyogo 650-0046, Japan Hideo Kasuga 3

Kanagawa Institute of Technology 1030 Shimo-ogino, Atsugi City, Kanagawa, 243-0292 Japan Mari Ueda 4

Kanagawa Institute of Technology 1030 Shimo-ogino, Atsugi City, Kanagawa, 243-0292 Japan

ABSTRACT Most of the sound quality evaluation methods for loudspeaker are based on physical evaluation and subjective evaluation. In the physical evaluation, we focus on the peak sound pressure and the rise speed. In the subjective evaluation, the sound quality is evaluated by the subjective evaluation experiment using the sound quality evaluation word. However, it’s difficult for people who don’t have professional knowledge in the field of acoustics to judge the performance of a loudspeaker from these evaluation methods. In addition, according to the engineers of a speaker manufacturer, the difference of the sound quality of the loudspeaker may affect the movement and motivation of the game. In this study, aims to evaluate the sound quality of loudspeaker objectively from a new point of view. In order to clarify whether the difference of the sound quality of the loudspeaker affects the movement, the movement analysis using OpenPose was carried out. As a result, it was found that the complicated motion was not likely to be affected by the difference of the sound quality of the loudspeaker. However, the results suggested that simple and familiar actions may be affected by the difference in loudspeaker sound quality.

1 s2185017@cco.kanagawa-it.ac.jp 2 r_matsuishi@toa.co.jp 3 kasuga@ic.kanagawa-it.ac.jp 4 m-ueda@ic.kanagawa-it.ac.jp

1. INTRODUCTION

Most sound quality evaluations of loudspeakers have so far been mainly based on physical evaluation and subjective evaluation by hearing. For example, physical evaluation focuses on sound pressure peaks and rise speed [1], and subjective evaluation often involves conducting sound quality subjective evaluation experiments using the ITU-R [2] step evaluation for evaluators. In addition, sound quality evaluation terms related to human sensitivity [3] are used in subjective evaluation experiments. However, it is difficult for people who do not have specialized knowledge in the field of acoustics to judge the sound quality of a loudspeaker from the aforementioned physical and subjective evaluations.

According to engineers of a speaker manufacturer, when preschool children danced the same dance with two different types of loudspeakers of different performance, the loudspeaker with better sound quality clearly made their movements more lively than the inexpensive loudspeaker. In addition, when loudspeakers with good sound quality were installed in a non-spectator game, the players' motivation was improved compared to that of inexpensive loudspeakers. The improvement of motivation may affect the movement.

Therefore, in this study, we clarify the influence of the difference in sound quality of loudspeakers on athletic movements by motion analysis using OpenPose, a posture estimation library. The objective of this study is to enable objective evaluation of loudspeaker sound quality by motion analysis, and to communicate the value of sound without the need for specialized knowledge. 2. Motion analysis with OpenPose

OpenPose, which is used for motion analysis in this study, is a posture estimation method that can extract a person's skeleton in two-dimensional coordinates from still or moving images. It can also detect the coordinates of the face, hands, and joints. The first advantage of using OpenPose is the ease of operation, even though the detection is performed in a two-dimensional coordinate plane. As shown in Figure 1, there is no need for participants to wear special suits or equipment, nor is there a need to use special cameras to capture images. There are also no restrictions on the location of filming, so there is no need to film in a dedicated studio. Data captured by commonly used monocular video cameras and smartphones can also be analyzed. Therefore, it is possible to film the experimental participants in a state as close as possible to the state in which they are exercising in their daily lives.

Figure 1: Captured image at OpenPose runtime.

3. New Acoustic Performance Index

Sound pressure-frequency characteristics are currently one of the most commonly used acoustic performance evaluation indices. However, since it is a two-dimensional measure in terms of sound pressure level and frequency, it is often illustrated or limited in scope when defining the target performance range. Therefore, acoustic knowledge is often required for those who evaluate the performance. In this study, a new LMH ratio is defined and a more concise rating of acoustic performance is considered.

3.1. Definition of LMH ratio

The LMH ratio defines the ratio of Low, Middle, and High energies. The calculation method determines the center frequency fc among the results of 1/3 oct. band analysis using pink noise. In this study, fc was set at 800 Hz. The arithmetic mean of the measurement results of the three bands including the ±1/3-oct. band (Equation 1) is determined as the energy representative of the midrange, with the frequency of the low frequency range being 0.1 × fc and that of the high frequency range being 10 × fc. Similarly for the bass and treble frequencies, the results of the 1/3 oct. band analysis of the three bands are arithmetically averaged (Equation 2 and 3), and the midrange is normalized by 0 dB, and the ratio of the low and high frequencies to the midrange (Equation 4) is used as an index for evaluating sound quality.

𝐸(𝑓 𝑐 − 1

3 𝑜𝑐𝑡)+𝐸(𝑓 𝑐 )+𝐸(𝑓 𝑐 + 1

3 𝑜𝑐𝑡)

𝐸𝑀=

3 , (1)

𝐸(0.1𝑓 𝑐 − 1

3 𝑜𝑐𝑡)+𝐸(0.1𝑓 𝑐 )+𝐸(0.1𝑓 𝑐 + 1

3 𝑜𝑐𝑡)

𝐸𝐿=

3 (2)

𝐸(10𝑓 𝑐 − 1

3 𝑜𝑐𝑡)+𝐸(10𝑓 𝑐 )+𝐸(10𝑓 𝑐 + 1

3 𝑜𝑐𝑡)

𝐸𝐻=

3 (3)

𝐿: 𝑀: 𝐻= 𝐸𝑀−𝐸𝐿: 𝐸𝑀: 𝐸𝑀−𝐸𝐻 (4)

4. Experiment Summary

In this experiment, we focused on dance movements in order to examine whether differences in motor behavior can be observed depending on the LMH ratio of the loudspeaker. Figure 2 shows the images taken while OpenPose was running. This paper reports the results of experiments 1) and 2).

Figure 2: Captured image at OpenPose runtime.

4.1. Experimental environment and experimental conditions for experiment 1)

Experiment 1) was conducted in the gymnasium of Kanagawa Institute of Technology. Participants were divided into two groups, A and B, with an average age of 21 years. 3 different acoustic patterns (a), (b), and (c) with different LMH ratios were used, and each group was filmed 3 times, for a total of 6 motion shots. The interval between them was 15 minutes. The position of the participant dancing in the center of the three groups was set so that the sound pressure level at the beginning of the song was 80 dB, and loudspeakers were set up at a distance of 4 m80 cm. The filming equipment was set up at a height of 1 m30 cm in front of the participants at a distance of 5 m30 cm from the participant

in the center, and slow-motion filming (1080p/240 fps) was performed using an iPhone XR. The sound sources and target dances used were "Koi" and "Koi Dance" by Gen Hoshino.

4.2. Analysis results of experiment 1)

In order to examine differences in movements based on the coordinates extracted from OpenPose, we decided to examine differences in movements based on the relative coordinates of the neck and limbs (coordinate values of the limbs - coordinate values of the neck) with the coordinate of the neck as the reference value. As an example, a scatter plot of the analysis results for one person is shown in Figure 3.

Figure 3: Analysis result of right wrist.

Figure 1 shows that no significant differences in behavior were obtained for any of the acoustic patterns. The shapes of the graphs were generally similar, and the maximum and minimum values of the coordinates were almost the same. The same result was obtained for the Y coordinate. Similar results were also obtained for other parts of the body, such as the left wrist and foot, and for other participants in the experiment. Therefore, it can be said that there were no significant differences in the movements in this experiment.

4.3. Experimental environment and experimental conditions for experiment 2)

Experiment 2) was conducted in a classroom at Kanagawa Institute of Technology. Participants were divided into two groups: A group of three and B group of two, alternating between the two groups. In this experiment, we analyzed the movements of radio calisthenics, in which simple repetitive movements are performed. The purpose of radio calisthenics is to improve physical fitness and to maintain and promote health, and it is a familiar exercise for Japanese people. It consists of various exercises such as stretching and arm twisting. The other experimental environment and conditions were the same as in experiment 1).

4.4. Analysis results of experiment 2) As in Experiment 1), in order to examine differences in movements based on the coordinates extracted from the OpenPose, we analyzed the relative coordinates of the neck and limbs (coordinates of the limbs - coordinates of the neck), using the coordinates of the neck as the reference value. Analysis of variance of the relative coordinates of the neck and limbs obtained from acoustic patterns (a), (b), and (c), respectively, revealed significant differences in back- stretching movements. The Bonferroni method was used to determine which pattern showed a significant difference between the two patterns. Figure 4 shows a scatter plot of the significantly different movements.

Table 1: Comparison by pattern Comparison Patterns P-value Significant difference Pattern1-Pattern2 0.005874126 * Pattern1-Pattern3 0.050850983 Pattern2-Pattern3 0.361785549 補正後α： 0.016666667

BIRO x eR —Patena Fe-h

Figure 4: X coordinates of stretching exercise

The scatter plots shown in Figure 4 indicate that there is a difference in the motion of each pattern in the hand turning movement. Especially around the 1000th frame, the differences in motion seem to be more pronounced. However, this difference in motion is not due to differences in the acoustic performance of the loudspeakers, but rather to habituation through repetition of the motion. 5. Conclusion

In this study, dance movements were analyzed using OpenPose in order to clarify whether differences in loudspeaker sound quality affect movement behavior. The results showed that in experiment 1), there were no significant differences in the movements of all participants in any of the patterns. One of the reasons for this is that the participants in this experiment had no experience of learning dance in earnest. Therefore, they concentrated their attention on dancing and were hardly affected by the difference in sound quality. In addition, we were able to confirm some differences in behavior in Experiment 2), but it was not due to differences in the acoustic performance of the loudspeakers, and may have been caused by other factors. As a future prospect, we plan to determine the participants of the experiment for each target. For example, we plan to limit the number of participants to those who have studied dance before, and we also plan to select participants for each age group, such as preschoolers, elementary school children, and the elderly. In this experiment, the same loudspeakers were used and the LMH ratio was changed to express the difference in acoustic performance. However, it was difficult to perceive the difference in acoustic performance in the actual experiment, and this may be one of the reasons why the behavior was not affected. Therefore, we will consider the possibility of expressing the difference in acoustic performance by changing the loudspeaker itself. 6. ACKNOWLEDGEMENTS

This research was supported by JSPS Grant-in-Aid for Scientific Research JP18K18625. We thank the authors for their kind attention.

7. REFERENCES

1. G.Kosugi et al., "Research and Development of Amplifier Intended for High Efficiency Energy

Radiation," Technical Report of the Institute of Image Information and Television Engineers, Vol. 33 No. 11, pp. 127-130, (2009). 2. Subjective assessment of sound quality,ITU-R BS.562-3 ， (1978) ． 3. M.Miyahara et al., "Survey analysis of evaluation words expressing sound quality," Journal of

the Acoustical Society of Japan, Vol. 52, No. 7, pp. 516-522, (1996).