The acoustic effects of temporal and spectral characteristics on traffic sound sensitivity

2.1 Objective measurement

Objective measurements were conducted in Hung Hom, Yau Tsim Mong District, where the percentage and number of residents exposed to noise levels exceeding 70 dB(A) are the highest in Hong Kong [10]. Hung Hom, a predominantly residential area in the southeast of the Kowloon Peninsula, connects three subway stations: Hung Hom, Ho Man Tin, and Whampoa. This area experiences heavy pedestrian and commuter traffic, with numerous heavy vehicles, light vehicles, and public buses on the roads. Therefore, four measurement points and a comparative point were selected to access the traffic sound levels at different road intersections, as shown in Figure 1. All measurement points are linked to the major road network, which has a high capacity of traffic flow. According to local authorities in 2022, the annual average daily traffic values are around 10,000, and the maximum road speed limit is 50 km/h. The surrounding buildings of all measured points are adjacent to high-rise residential structures (over 20 floors), and detailed information of all measured points was provided in Appendix A (see Table 6).

Figure 1:

The measurement locations: measurement point 5 is the comparative point. (The width of single traffic lane is 3.5 m).

The Brüel & Kjær type 2,270 sound level meter and the Brüel & Kjær type 4,189 microphone were used for the measurement. The calibration test was conducted by the Brüel & Kjær type 4,231 sound calibrator, with a stated accuracy of ±0.2 dB. The calibration was performed before and after each measured recordings, and the maximum permissible deviation for accepting a calibration check was set to ±0.5 dB. The result showed that the value of the microphone was 46.9 mV/Pa, which basically matched the value of the open-circuit value. As shown in Figure 1, the sound level meter was located at the roadside or the pedestrian refuge island near the waiting line, and it was also close to the sound source of traffic light signals. The height of the sound level meter was set to 1.5 m (±0.2 m), and the horizontal location was equal to the minimum distance from the nearest traffic light signals to the pedestrian waiting area (described by the traffic light distance in Table 2). Before the formal measurements, the pre-inspection visits to all intersections have been conducted to avoid the non-traffic sources in the measured area (e.g., construction noise or human event sounds).

Sound recordings and the root-mean square sound pressure of the traffic noise were recorded to create the audio WAV files with a 48 kHz sample rate, and noise data ranging from 7 to 10 min was recorded at each measurement point. The red-light and green-light signals cycled repeatedly, and the data recordings were divided into separate segments according to the duration of the red-light or green-light signals. The signal transition from red-light to green-light signals is instantaneous, but there is a flashing period (about 5 s) from green-light to red-light signals. The segments that lasted less than 20 s were eliminated for statistical stability because the 10th percentile of segment duration (red-light signals) was 18.2 s. The flashing green-light signals were also ignored in this study due to the short duration.

Tests were conducted separately during the traffic peak time (6 p.m.–9 p.m. on March, 27, 2024) and the off-peak time (1 a.m.–4 a.m. on April, 21, 2024). The measurements were carried out in conditions without any precipitation. The temperature was between 20 and 25°, the relative humidity was less than 90 %, and the wind speed did not exceed 2 m/s (data were obtained in real time from the Hong Kong Observatory Application). Moreover, the pre-measurement found that the perceived characteristics of the traffic-light signal were only minimally affected without a windshield. Therefore, the windshields’ protection was not used on the sound level meter’s microphone due to the low wind speed. To minimize the effect of wind interference, the mel spectrograms of all segments were examined to reduce the potential wind noise.

All non-traffic events were tagged in the measurement log, and clear data exclusion criteria were established: any tagged segment of impulsive non-traffic noise, including high-level human event sounds and car horns, was excluded from the final analysis dataset. All available segments of audio WAV files were analyzed by the acoustic toolbox in MATLAB 2024b, and the psychoacoustic indicators can be obtained by the built-in MATLAB program. The detailed parameters of calculations and acquisition settings were provided in Appendix A (see Table 5). The A-frequency weighting and the fast time constant was set to the SPL indicators of L _Aeq and L _max. All the data from statistical analyses were coded and analyzed by the commercial package SPSS, version 27.0.

2.2 Psychoacoustic indicators

A range of psychoacoustic indicators will cover the traditional indicators (L _Aeq and L _Amax) and the psychoacoustic indicators (N, S, FS, R, T), and the temporal characteristics for all indicators can be recorded by the time-varying curves [23]. The concept of loudness (N) is used to calculate the sound intensity for human perception. The calculation of the loudness model is based on the ISO 532-1 standard [41], and the equation can be expressed as follows:

Specific loudness (N′) is expressed as a function of the critical band rate (the unit is sone/Bark). The equation of specific loudness can be expressed as:

In this equation, E _TQ is the excitation at the threshold in quiet and E ₀ is the excitation of the noise corresponding to the reference intensity I ₀ = 10⁻¹² W/m².

The indicator sharpness (S) is strongly correlated with the spectral distribution and the spectral centroid (SC). Spectral centroid describes the distribution of power at different frequencies, and the sharpness model (DIN 45692) is defined by an analogous equation [42]:

where g(z) is the function of the additional factor g on the critical-band rate. The value of g(z) is equal to 1 when the critical-band rates are <16 Bark (approximately 3,150 Hz).

Fluctuation strength (FS) is a temporal indicator related to frequency modulation (FM) and amplitude modulation (AM). Fluctuation strength describes how the sound fluctuates perception over time, reaching its maximum near modulation frequency of 4 Hz and decreasing at higher modulation frequencies. Roughness (R) indicates the spectral perception of stimuli with AM or FM rates, and this value reaches its maximum when the frequency modulation is close to 70 Hz. These indicators based on the Zwicker’s model can be given as follows [23]:

where, ΔL is the temporal masking depth related to the time-varying amplitude, and ΔL _E is the masking depth modified by the 1 kHz pure tone. The integral range of ΔL and ΔL _E can cover the entire critical-band rate. f _mod is the modulation frequency, and the unit is Hz.

The rhythmic event can be calculated by the temporal pattern in the loudness curves, and it has a significant correlation with fluctuation strength (FS). Sound signals with a quick rise in the loudness curves will deviate from the physically temporal perception, and a subjectively rhythmic perception can be produced to adapt the human auditory system [23]. The model of the rhythmic event can be defined by the time-varying loudness within a period of time [43]:

where, N _M denotes the loudness curves of the absolute maximum during a total period. N _max and N _min are the loudness curves of the relative maximum and minimum during a segmental period. Δt is the minimum time gap between two rhythmic events, and its value needs to be greater than the minimum perceptible time gap (120 ms). The whole sound signal can be divided into several segments according to time-varying loudness curves. A single rhythmic event can be determined when the above three conditions are met in each segment, as shown in Figure 2. Finally, the total number of rhythmic events over a period of time can be calculated.

Figure 2:

The time-varying loudness curves for showing the example of three calculated rhythmic events.

The indicators of Tonality (T), Tone Level (TL), and Tone-to-noise ratio (TNR) are also considered to describe the tone and pitch of the spectral characteristic, and there are significant correlations among these three indicators [31]. According to the ECMA standard, TNR is the tonal measurement related to the background sound level, and it will determine how clear or noticeable an acoustic event is to the listener [44]. The relationship between the tonal sound and background noise is more comprehensively considered in the TNR indicator, and it is suitable for traffic light signals. A discrete tone will be judged as prominent according to the prominence ratio (prominence tone) method if the TNR value is higher than 9 dB.

Previous studies have shown that a fluctuating sound recording does not correspond to the average value of loudness curves but rather to a value near the maximum [23]. Traffic light signals differ from general traffic noise due to their dramatic changes over time. Therefore, the percentile loudness indicator N ₅ was used to evaluate traffic noise, where N ₅ represents the loudness reached or exceeded within 5 % of the measured time. Furthermore, the aforementioned indicators can be integrated into a comprehensive indicator that describes sound sensitivity and perception. In Zwicker’s model, Psychoacoustic annoyance (PA) can quantitatively analyze annoyance ratings in field measurements [23]. Besides intensity indicators like loudness, psychoacoustic annoyance also depends on temporal and spectral characteristics. The equation can be expressed as follows:

Where, the value of ω _S is 0 if the S is <1.75. The unit of the PA is sone, which is consistent with the unit of loudness. The psychoacoustic indicators N ₅, S, F and R are considered in the PA model, whose units are sone, acum, vacil, asper respectively. To sum up, all psychoacoustic indicators with methods are summarized in Table 1.

2.3 Subjective listening tests

The listening tests were conducted in a quiet room (length 4.2 m × width 3.2 m × height 2.8 m) with a background noise level of 39 dB (A). The listener sat in a chair in front of a table, and the indoor air quality is good. Therefore, the participants can concentrate on the recorded segments and give the effective answers. The recorded segments were played back using an acoustic system that included headphones (Sennheiser HD 600), a high-quality sound card (Focusrite Scarlett Solo 3), and a computer. No additional headphone equalization or HRTF correction was applied to preserve the natural response of the Sennheiser HD 600. The linearity of playback chain (Focusrite Scarlett Solo 3 with HD 600) was verified by the given total harmonic distortion plus noise (THD + N < 0.1 % at 1 kHz, 97 dB SPL). All segments were sourced from the segments mentioned in Section 2.1, and the B&K 4101-B in-ear microphone was used to calibrate the sound pressure levels from the headphones. Due to the lower sound pressure levels (−4.4 dB) of off-peak segments in Table 3, the peak and off-peak segments were calibrated using a standard sound source at 60 dB and 65 dB, respectively.

The experimental procedure and the following instructions were given to participants: “You need to evaluate eight recorded segments of urban sound environments. Imagine yourself walking at road intersections and rating your annoyance on a continuous scale ranging from −3 to 3 (7 scales). You can hear each segment three times, and you will not be able to go back to the previous segments if you validate your answer.” After hearing a background traffic sound (without traffic light signals), eight sound segments were played one by one in random order, and each segment lasted 40–60 s. The tests for each participant lasted approximately 15–20 min. Total 15 participants (13 male and two female), with normal hearing and stable emotion, participated the tests. All participants were aged between 20 and 30 years to control the similar auditory acuity. The participants were asked to score their annoyance, sensitivity, and perceptions by different semantic adjective pairs in Table 2. The answer scale in Table 2 refers to the three-dimension model (EPA model) [25]. The EPA model evaluated the human perceptions about the general judgement, the energy content, and the temporal and spectral content of sound, respectively [45]. Moreover, several additional adjective pairs, including the pairs recommended by the standard ISO 12913-3, were also added to consider the signal-related perception [17], [46], [47], [48], [49], [50].

3.1 Temporal characteristics

3.1.1 Time-varying curves of measured indicators

There were 29 and 27 available segments of red-light signals and green-light signals, respectively, as shown in Table 3. The off-peak segments at measurement point 4 were excluded due to the lower volume of the traffic-light signals. Consecutive segments at measurement point 2 were selected to specifically highlight the temporal trend in the objective results. Four time-varying curves of measured indicators at measurement point 2 for two consecutive segments were presented in Figure 3, and the red lines represent the signal transition from the red signal to the green signal. The time-varying curves of SPL and N in Figure 3(a and b) fluctuated significantly during the red-light signals but remained relatively steady during the green light.

Figure 3:

The time-varying curves of two consecutive measured segments (peak time measured at the point 2): (a) sound pressure level; (b) loudness; (c) fluctuation strength; (d) sharpness.

The difference in traffic noise between peak and off-peak times was also significant in Figures 3 and 4. During the off-peak times, the time-varying curves of SPL and N remained relatively flat due to the reduced traffic light volume and vehicle noise. The temporal envelope of red-light signals showed sharp fluctuations, particularly in loudness. In Figure 4(a), the different temporal envelopes of red-light and green-light signals disappeared, but this difference persisted in Figure 4(b). It indicated that the patterns of the time-varying N differ significantly between the red-light and green-light signals. Comparatively, the actual perception of people (N result) is more sensitive than the physical energy change (SPL result) for traffic light signals [23].

Figure 4:

The time-varying curves of two consecutive measured segments (off-peak time measured at the point 2): (a) sound pressure level; (b) loudness; (c) fluctuation strength; (d) sharpness.

Figures 3(c) and 4(c) illustrated the time-varying FS curves of typical results. During the green-light signals, FS values peaked at approximately 0.3 vacil with a steady modulation frequency (f _mod) of 12.8 Hz. In contrast, during the red-light signals, FS values fluctuated and dropped to around 0.15 vacil. It indicated a clear difference in the temporal patterns during the signal transition, with an inverse relationship between N and FS in Figure 3(b) and (c). The temporal envelope of time-varying S curves in Figures 3(d) and 4(d) showed the steady patterns during the red light, with the disordered patterns during the green light.

While vehicles passed by in the adjacent lane (referenced at the 65-s mark in Figure 4(c)), there was a clear decrease in FS curves during the green-light signals. The lower-frequency range of the traffic light signals often gets masked by the more prominent low-frequency vehicle noise, which mirrors the rising trends of traffic noise observed in prior studies [18], 21]. This interference can lead to a shift in f _mod, resulting in a decrease in FS. Moreover, a modulation frequency of f _mod = 12.8 Hz was distinctive of traffic light signals. As indicated by the comparative data points, f _mod tended to be unstable at the intersections without traffic light signals, and FS values also diminished to a lower level as a result.

The difference in time-varying R observed in Figure 5 between the red-light and green-light signals primarily stemmed from the sound sources present (vehicle noise and traffic light signals), and f _mod was consistently around 12.8 Hz during the green-light signals. However, the majority of f _mod hovered within the 80–100 Hz range during the red-light signals, with only a few segments retaining the 12.8 Hz frequency. In contrast to FS values, vehicle noise can influence the perception of R, with f _mod typically rising within the low-frequency range (80–100 Hz). During the red-light signals, the increase of vehicle noise can overshadow the traffic light signals, and the variation in f _mod contributed to a slightly higher R during the red-light signals [22].

Figure 5:

The time-varying curves of roughness (measured at the point 2): (a) green-light signal; (b) red-light signal.

3.1.2 Analysis of rhythmic events

The concept of beats per minute (BPM) proves useful in quantifying the frequency of uniform fluctuations per minute in a stable signal [51], 52]. As shown in Figure 3(b and c), the pure traffic light signals (without random vehicle noise) presented the regular patterns in the loudness time-varying curves. Notably, the BPM of the pure green-light signals, set at 608 BPM, was eight times higher than that of pure red-light signals (BPM = 76) across all measurement points. Therefore, the concept of rhythmic events per minute (REPM) was defined to analyze the measured segments on a BPM scale. These rhythmic events were computed and categorized using the MATLAB function module. However, the REPM values for all green-light signals were disregarded due to the brief intervals. The intervals of green-light signals were less than the minimum perceptible time gap (120 ms) in Eq. (8).

The BPM value of 76 for red-light signals, as depicted in Figure 6, served as a notable threshold positioned within the upper interquartile range (IQR) of peak-time and overall segments, with the upper limit around BPM = 76 even during the off-peak hours. The dynamically changing time-varying loudness (N) and maximum loudness (N _max) derived from vehicle noise ultimately contribute to REPM values generally falling below BPM. Consequently, REPM was associated with stable red-light signals and lower sensitivity to vehicle noise. The presence of vehicle noise introduced instability, with a decrease of REPM. Nevertheless, the distinguishable rhythmic perceptions remained predominantly induced by the traffic light signals.

Figure 6:

The calculated rhythmic event per minute (REPM) for all red-light segments.

3.2 Spectral characteristics

The mel spectrogram in Figure 7 illustrated the specific loudness (N′) at the critical band rate scale, emphasizing the spectral characteristics over time. The 10-s segments were selected to highlight the signal transition within a short duration. During the red-light signals, these frequencies appeared as a series of regular peaks, indicating consistent oscillations throughout the period, visualized as vertical-line patterns in Figure 7(b) and (d). In contrast, the green-light signals generated a more frequent vertical-line pattern, with a multiple of 8, aligning with the findings in Section 3.1.2. During the signal transition in Figure 7(c), both the red-light signals and the green-light signals produced distinct frequencies at 5, 13, 16, and 20 Bark. Moreover, a repeated measurement of the covered windscreen has verified that wind interference does not influence the distinct frequencies of traffic signals.

Figure 7:

The mel spectrogram of the specific loudness (N′): (a–c) vehicle passing and signal transition at measured point 2; (d and e) vehicle passing at measured point 1.

Due to the lack of concurrent traffic data, Figure 7(a) and (e) selected the spectrograms for different traffic-light signals to visually illustrate the impact of vehicle noise. It indicated that the process of a vehicle passing through the intersection typically lasts no more than 5 s. During this time, N′ values around 2 Bark increased significantly due to the vehicle, and a series of irregular peaks within the 0–10 Bark range can be observed in the spectrograms. Portions of the traffic signal’s frequencies at 5 Bark were masked by random vehicle noise, but the high-frequency components at 16 and 20 Bark remained unaffected. Previous investigation of traffic noise consistently indicated that the high-frequency proportion above 3,000 Hz of vehicle noise was extremely low [21].

In terms of the total duration, Figure 8 illustrated the mean N′ values of different traffic light signals across critical-band rates for all segments. As shown in Figure 8(a and b), the N′ values of peak time were significantly higher than that of off-peak time in the range of the entire 0–24 Bark. Meanwhile, N′ around 2 Bark and 16 Bark showed an obvious elevation during the peak time compared to the off-peak time, but the elevations of N′ around 5 Bark and 20 bark were not as pronounced as at two bark and 16 bark. The N′ values around 2 Bark were slightly higher during the red-light signals due to the increased vehicle noise. However, the red-light and green-light signals were all concentrated around 16 Bark in spectral characteristics.

Figure 8:

The mean specific loudness for all segments: (a) peak time; (b) off-peak. (The red and green lines denote the red and green-light signals respectively).

In Figure 9(a), the tone-to-noise ratio (TNR) highlighted a consistent prominent frequency of approximately 3,270 Hz within traffic noise, corresponding to N′ at 16 Bark. The blue peak in Figure 9(a) also appeared at various characteristic frequencies at at 2,200 Hz (12 Bark), 6,500 Hz (20 Bark), and 10,000 Hz (23 Bark). The characteristic frequency range was consistent with the mel spectrogram results in Figure 7. Nevertheless, TNR values exhibited significant variation across different segments, with only a minority meeting the criteria for prominence tone as depicted in Figure 9(b). The reason is that the acoustic design of the traffic signal source also considered the repetition rate (temporal) characteristics [18], 19]. Moreover, the traffic light signals introduced a distinct tonal presence or spectral pattern, with high frequencies (around 16 Bark) being further influenced by vehicle noise in spectral perception [21]. Consequently, only a few segments characterized by low-level random vehicle noise surpassed the prominence tone threshold at 16 Bark.

Figure 9:

Tone-to-noise ratio results: (a) tone-to-noise ratio in frequency spectrum; (b) results of all segments with the prominence tone threshold.

4.1 Statistical description for the measured indicators

According to the standard of traffic light signals (ISO 23600:2007), the sound pressure level of all traffic light signals should be maintained between +5 dB and +10 dB relative to the ambient noise level [53]. However, the perceptual differences were not explained between red-light and green-light signals due to the lack of psychoacoustic indicators. Table 4 presented the statistical outcomes for both psychoacoustic and traditional indicators across all segments in the measurement. The measurement revealed that the mean values of N _eq, N ₅, N _max, L _Aeq, and L _max for both red-light and green-light signals were nearly identical. During the green-light signals, the mean and median values of most indicators were slightly higher than during the red-light signals.

The Shapiro-Wilk tests result indicated that two indicators (R and FS) do not follow a normal distribution (p < 0.05). The nonparametric tests (Mann-Whitney U test) were conducted to examine whether the indicators have significant differences between the green-light signals and the red-light signals. The null hypothesis was that there are no statistically significant differences in the subjective data for individual questions, and the effect size r was calculated by rank-biserial correlation [54]. Two indicators, S (p = 0.011, r = 0.42) and FS (p < 0.001, r = 0.91), showed significant differences for different traffic lights. However, the statistical distributions of S were still similar for different traffic lights in Figure 10(a) due to the same standard deviation and approximate mean. In contrast, the distributions of FS varied significantly for different traffic lights in Figure 10(b), with the distribution during the green-light signals being more dispersed than during the red-light signals.

Figure 10:

The probability distributions of significantly different indicators for all segments: (a) sharpness; (b) fluctuation strength.

4.2 Statistical description for subjective listening tests

A total of 24 segments from the aforementioned segments were further evaluated for the listening tests, with 12 segments from red-light signals and 12 segments from green-light signals. A questionnaire for one participant included six peak segments and two off-peak segments to consider the complex and random vehicle noise. 15 participants with 60 valid questionnaires were collected for the red and green light signals, respectively (sample size, n = 120). The reliability statistics of the questionnaires’ results showed a good level of subjective results, with the Cronbach’s Alpha coefficient at 0.885 [42].

Figure 11 presented the mean scores for various questions, categorized by the types of traffic lights. It can be seen that all results in the questionnaire had a positive value, so the acoustic environment at the intersections was annoying and unpleasant for participants. Question two exhibited a similar sensitivity to green-light signals (M = 1.60, SD = 1.21) compared to the red-light signals (M = 1.38, SD = 1.49), without the statistical significance (p = 0.375). Furthermore, Question 3 (sensitivity to vehicle noise) was slightly higher during the red-light signals (M = 1.85, SD = 1.51) than the green-light (M = 1.70, SD = 1.61), but the result of Mann-Whitney U tests showed no significant difference between the two groups (p = 0.714). However, the score for most questions in the green-light signals was higher than that of the red-light signals, and Question 4 (sensitivity to traffic light) for the green-light signals (M = 1.83, SD = 1.52) was significantly greater than that of the red-light signals (M = 1.30, SD = 1.46, p = 0.038).

Figure 11:

Mean scores and standard deviation of the subjective listening tests: questions 2–5 denote the sensitivities to different sounds, questions 6–17 denote the different perceptions.

The Shapiro-Wilk tests indicated that all subjective data do not follow a normal distribution (p < 0.05). Consequently, the Mann-Whitney U tests were conducted to assess the statistical significance of the subjective data in Figure 11. The perceptions of tense/relaxed (p = 0.032, r = 0.22), fluctuant/steady (p < 0.01, r = 0.48), and fast/slow tempo (p < 0.01, r = 0.71) rejected the null hypothesis, as well as the sensitivity to traffic light signals (p = 0.038, r = 0.22). Based on the aforementioned perceptions, participants in the listening tests can significantly distinguish traffic light signals during the signal transition. Prolonged exposure (over 30 s) to the green-light signals was associated with higher reported sensitivity and perception scores compared to the red-light signals.

4.3 Correlation analysis

A Spearman correlation analysis was conducted to explore the relationship between the subjective data and measured indicators. Figure 12 presented heatmaps showing the correlations between the measured indicators and the subjective data, and the heatmaps can be explained by three parts. Firstly, vehicle noise was highly correlated with the measured indicators N _eq, N ₅, N _max, L _Aeq, L _max, R, PA (p < 0.01), and S (p < 0.05). The perceptions of tense/relaxed and loud/quiet showed the highest correlation with the intensity indicators N _eq, N ₅, and L _Aeq. However, the correlations between vehicle noise and several other indicators (FS, TNR, and REPM) were relatively low, at −0.07, 0.17, and −0.18, respectively.

Figure 12:

The heatmaps of the correlations between the measured indicators and the questions of the listening tests (REPM indicator only considered red-light signals).

Secondly, the sensitivity to traffic light signals was correlated with all measured indicators except for R in Figure 12. The correlations of S, FS, TNR, and REPM with traffic light signals were significantly higher than with vehicle noise. The correlation coefficient between FS and the slow/fast tempo was the highest at 0.48. Therefore, a higher perception of a fast tempo was associated with higher FS values, which was in turn linked to higher sensitivity to traffic light signals. The increased perception of fast tempo was also consistent with the higher FS values during the green-light signals. Meanwhile, FS showed almost no correlation with vehicle noise and its corresponding perceptions (<0.10), and it was only correlated with the signal transition. The REPM had a correlation with sensitivity to traffic light (0.21) and significantly negative correlations with irregular/regular (−0.32), heavy/light (−0.28), and sharp/dull (−0.27).

Another study also identified elevated FS values in relation to construction noise generated by pile drivers and breaker machinery, with the time-varying loudness curves showing patterns similar to those of red-light signals [55]. However, several studies focused on the vehicle noise or train noise have not identified any sound sources directly associated with FS [30], 31]. Moreover, the subjective listening tests employed tempo-related descriptors, including non-stationary/stationary, continuous/intermittent, and non-periodic/periodic, which were significantly associated with FS values [55]. The findings suggested that variations in FS may serve as a reliable independent indicator for evaluating the signal transition of traffic lights. The standard ISO 23600-2007 did not provide a specific description of the signal transition [53]. In summary, the temporal indicators (FS and REPM) effectively captured temporal variance over time, which was significantly correlated with perceived rapid tempo and regularity.

Finally, S also shows a high correlation with two temporal-related perceptions, likely due to the strong correlation (0.55) between the measured indicators FS and S. Additionally, while the correlation between TNR and sensitivity to traffic noise was high (0.34), its correlation with multiple perceptions in the questionnaire was relatively weak. Another study also identified the frequency of traffic noise and signal sounds, which all widely ranged from 400 to 3,000 Hz [21]. The overlapped spectral characteristics of traffic signal sounds are highly susceptible to interference from random traffic noise. Changing spectral indicators (S, R, and TNR) can significantly affect sensitivity, but temporal indicators provided clearer perception for distinguishing between the red-light and green-light signals [20].