An evaluation of discomfort reduction based on auditory masking for railway brake sounds

Transcription

1 PROCEEDINGS of the 22 nd International Congress on Acoustics Signal Processing in Acoustics: Paper ICA An evaluation of discomfort reduction based on auditory masking for railway brake sounds Sayaka Okayasu (a), Takahiro Fukumori (b), Masato Nakayama (c), Takanobu Nishiura (d) (a) Graduate School of Information Science and Engineering, Ritsumeikan University, Japan, (b,c,d) College of Information Science and Engineering, Ritsumeikan University, Japan, Abstract Railway brake sound in slowing railway vehicle causes noise problems in a station yard. Conventional noise reduction methods for railway brake sound have been proposed on basis of the improvement of brake mechanism. However, these methods have the insufficient performance of the noise reduction for railway brake sound and the railway brake sound still gives passengers discomfort feeling. In this paper, we focus on active control for reducing discomfort feeling of railway brake sound. We have previously proposed a discomfort reduction based on the auditory masking for a stationary sound. In this method, discomfort reduction is realized by emitting the noise control masker signal with the secondary loudspeaker without updating the estimated noise. However, the performance of discomfort reduction is insufficient for the railway brake sound because the railway brake sound is non-stationary noise. In this paper, we therefore propose a new discomfort reduction method based on auditory masking with the active noise estimation for the railway brake sound. We carried out subjective evaluation experiment with the actual railway brake sound. As a result, we confirmed the effectiveness of the proposed method. Keywords: Railway brake sound, Auditory masking, Discomfort reduction, Masker signal

2 An evaluation of discomfort reduction based on auditory masking for railway brake sounds. 1 Introduction In slowing old-type railway vehicle, the railway brake sound causes noise problems in a station yard. In the conventional researches, various methods for noise reduction have been proposed. Conventional noise reduction methods for the railway brake sound have been proposed on basis of the improvement of brake mechanism [1][2]. However, these methods have the insufficient performance of the noise reduction for the railway brake sound. The railway brake sound still gives passengers discomfort feeling. Therefore, we focus on discomfort reduction of that without improving brake mechanism. Active noise control (ANC) [3] has been proposed in order to reduce noise pressure by emitting opposite phase signal with the secondary loudspeaker. ANC is effective as the noise cancelling headphone. However, it is difficult to control large space such as station yard because the method controls a small point space. Also, the discomfort reduction based on auditory masking has been proposed [4][5]. Auditory masking is the phenomenon that a sound is difficult to be heard being covered by other sound caused by auditory characteristic of human [6]. This method can reduce discomfort feeling by emitting the masker signal with secondary loudspeaker. This method has been utilized for reducing discomfort of an airconditioner noise, a turbine sound, a dental treatment sound because these sounds have a stationary noise with spectral peaks. On the other hand, the railway brake sounds have a non-stationary noise with spectral peaks. This suggests that it is difficult to introduce this method to railway brake sounds. In this paper, we therefore propose a new method of discomfort reduction based on auditory masking with the adaptive noise estimation for the railway brake sound. The proposed method analyzes the non-stationary noise of the railway brake sound in frame by frame. Also, the proposed method designs the masker signal by using the estimated noise at real time. The masker signal is emitted with the secondary loudspeaker. Finally, we carry out subjective evaluation experiment with the actual railway brake sound to confirm the effectiveness of the proposed method. 2 Railway brake sound Old-type railway vehicle emits high-frequency noise in slowing in station yard. This high-frequency noise is called as the railway brake sound. It gives discomfort feeling many people in station yard. Figure 1 shows the spectrogram of the railway brake sound. The vertical axis shows the frequency, and the horizontal axis shows the time. The shades of collar shows the log power. This railway brake sound is recorded in Kansai main line operated by the West Japan Railway Company as shown in Fig. 2. 2

3 Figure 1: Spectrogram of the railway brake sound. Figure 2: Photograph of the railway vehicle. From Fig. 1, the railway brake sound has multiple spectral peaks from 2 to 12 khz. Noise with spectral peaks raises discomfort feeling [7]. Also we comfirmed that multiple spectral peaks and those sound pressure levels vary in the time. This result indicates that the railway brake sound is non-stationary noise. 3 Conventional discomfort reduction based on auditory masking for stationary noise We have proposed the discomfort reduction method based on auditory masking for the stationary sound as the conventional method. The conventional method can reduce discomfort feeling by emitting the masker signal which can mask the spectral peak with secondary loudspeaker. Spectral peak is discomfort component of noise. Figure 3 shows the flow of the conventional method. As shown in Fig. 3, the conventional method consists of two steps. Step 1) Detect the single spectral peak The conventional method assumes that the stationary noise with the single spectral peak is observed with microphone. In this step, the power spectrum of the noise is calculated from the observed noise using Fourier transform, and the spectral envelope is calculated from the power spectrum using smoothing processing. Also, to determine the single spectral peak, the local standard deviation is calculated as follow: V i F = P i μ i F σ i F (1) where i is frequency, F is frequency bandwidth for calculation, V F i is the local standard deviation, P i is the spectral envelop and μ F i is the average of P i in F [Hz] that centers on i [Hz], and σ F i is the standard deviation of that. V F i represents prominent value of the power of i [Hz] compared with that of the frequency width around i [Hz]. The frequency of the single spectral peak is calculated as follows: 3

4 Figure 3: The flow of the conventional method. Figure 4: Overview of the spectra of the noise and the masker signal. f = argmax(v F i ), (2) i where f is frequency of the single spectral peak which has the maximum value of local standard deviation. Step 2) Design the masker signal Figure 4 shows overview of the spectra of the noise and the masker signal. As shown in Fig. 4, the critical band [8] is very important in order to design the masker signal. Critical band is minimum frequency range that can mask center frequency of pure tone. Critical band is fomulated by 4

5 Figure 5: The flow of the proposed method. f p CB(f p ) = ( ( 1000 ) ), (3) where CB(f p ) is the critical band, and f p is the frequency of the single spectral peak. The band pass filter with bandwidth of CB(f p ) is designed as follows: h fp (t) = 2 f p,e1 sinc (2πtf p,e1 ) 2 f p,e2 sinc (2πtf p,e2 ), (4) f p,e1 = f p + CB(f p), 2 (5) f p,e2 = f p CB(f p), (6) 2 h fp (t) is the band pass filter which is used to design masker signal. f p,e1 and f p,e2 are upper and lower limit frequencies of band pass filter. To design bandlimited noise as the masker signal, the masker sound source is filtered by the band pass filter. White noise, pink noise and running water sound are generally utilized as the original sound sources of masker signal. However, the railway brake sound has multiple spectral peaks. Thus, in this paper, we design the masker signals for largest spectral peaks from 2 khz to 12 khz. 4 Suggestion of discomfort reduction using adaptive noise estimation for non-stationary noise The proposed method assumes that the non-stationary noise such as railway brake sound with multiple spectral peaks is observed with microphone. In the conventional discomfort reduction 5

6 Figure 6: The determination overview for estimated spectral peaks. method cannot reduce discomfort feeling insufficiently for non-stationary noise because designed masker signal is difficult to follow the time variation. In this paper, we propose the new discomfort reduction method based on auditory masking for the non-stationary noise such as the railway brake sound. Figure 5 shows the flow of the proposed method. As shown in Fig. 5, the proposed method consists of three steps as follows. Step 1) Detect multiple spectral peaks in current frame The railway brake sound has multiple spectral peaks which vary in the time. That is to say, it is necessary to detect multiple peaks in frame by frame. The local deviation is calclated by Eq (1) as same as the conventional method. The spectral peaks are detected by using the threshold of local standard deviation. To detect the spectral peaks, the threshold was determined in the preliminary experiment. If critical bands of detected spectral peaks are overlapped, the proposed method employs the frequency of spectral peak which has larger local standard deviation. The frequencies of spectral peaks are recorded in all frames. Step 2) Estimate the frequencies of multiple spectral peaks using the adaptive noise estimation Figure 6 shows the determination overview for estimated spectral peaks. In Fig.6, the vertical axis shows the frequency and the horizontal axis shows the time. As shown in Fig. 6, the masker signal is estimated according to spectral peaks in current and several previous frames. We define the range that combines current and several previous frames as the reference frame. The number of the reference frame was determined in the preliminary experiment. In Fig. 6, circles show the spectral peaks which are detected in Step 1. Squares show the estimated spectral peaks which are used to design masker signal. Dashed line represents the overlapping area of critical band in 6

7 the reference frame. To adaptively estimate the noise with spectral peaks, the spectral peaks which are detected in the reference frame are employed as estimated spectral peaks. This is because the spectral peaks of railway brake sound intermittently arise at same frequencies. The detected spectral peak in the closest frame to head frame in the reference frame employed as a estimated spectral peak when critical bands of spectral peaks in the reference frame are overlapped. To design the masker signal, the gains for bandlimited noises are detemined by using the adaptive estimation with the powers of estimated spectral peaks. The gains for bandlimited noises are calculated by using Segmental Peak Signal-to-Noise Ratio(SP_SNR) as follows: M(f p, n) = 10 α(f p,n), (7) L 1 α(f p, n) = 1 L N(f p, (n l)) l=0 SP_SNR 10, (8) where n is the index of frame, f p is the estimated spectral peak, L is the total number of referenced frame, N(f p, (n l)) represents log power of f p in (n l)th frame of noise, M(f p, n) represents log power of f p in n th frame of the masker signal. The value of SP_SNR was decided in the preliminary experiment. Step 3) Design the masker signal with the estimated frequencies of multiple spectral peaks By using the frequencies of the estimated spectral peaks and the estimated gains in Eq. (7), the masker signal is designed as follows: I masker(t, n) = M(f i, n) s fi (nt + t), i=1 I = M(f i, n) w(t) h fi (t), (0 t < T) (9) i=1 where i is the index of estimated spectral peaks, t is the time index in frame, T is the total number of sample in one frame, n is the index of frame, I is the total number of estimated spectral peaks, f i is the i th estimated spectral peak, masker(t, n) represents masker signal, s fi (nt + t) represents masker signal for f i, w(t) is the original sound source of masker signal, h fi (t) is the band-pass-filter calculated by Eq. (4). In the proposed method, the designed masker signal is emitted with the secondary loudspeaker. 5 Subjective evaluation experiment We carried out a subjective evaluation experiment to confirm the effectiveness of the proposed method. 7

8 (a) The masker signal with white noise (b) The masker signal with pink noise Figure 7: The result of subjective evaluation experiment. Table 1: Condition of subjective evaluation experiment. Environment (A weighted sound level) Headphone Subjects Sampling Frequency / Quantization Number of sound sources Reference frame width Frame width / Frame shift SP_SNR Experimental criteria Office (38.5 [db]) SONY,MDR-CD900ST 7 [people] 48 [khz] / 16 [bit] Evaluation sound sources:5 -The original railway brake sound -The railway brake sounds processed by -the conventional method with white noise -the proposed method with white noise -the conventional method with pink noise -the proposed method with pink noise 2.0 [sec.] 0.5 [sec.] / 0.25[sec.] White noise: 5 [db] Pink noise: 0 [db] Comparison with reference sound 1:Discomfort 2:Quite discomfort 0:Neutral -1:Little discomfort -2:Not discomfort 5.1 Experimental condition We carried out a subjective experiment to evaluate value of discomfort reduction of railway brake sound using the proposed method. Table 1 shows the condition of the experiment. 8

9 In this experiment, the proposed method was compared with the conventional one by using the value of discomfort feeling. The conventional method emits stationary masker signal to mask the single spectral peak of railway brake sound. On the other hand, the proposed method emits nonstationary masker signal which is designed at regular intervals to mask the multiple spectral peaks of railway brake sound. We compare the value of discomfort feeling in a round-robin of the railway brake sound because Scheffe s paired comparison is used as a configuration method of evaluation criteria [9]. 5.2 Result and discussion of subjective evaluation experiment Figure 7 shows a result of the subjective evaluation experiment. The vertical axis shows the value of discomfort feeling. We confirmed sufficiently discomfort reduction using the masker signal which is designed in frame by frame. As a result, the value of discomfort feeling was decreased by about 1.0 compared with that of the original railway brake sound and that of the proposed method. The experimental criteria of the proposed method is lower than that of the original railway brake sound and the conventional method. There are reasons that masker signal which is designed by using the proposed method could sufficiently reduce discomfort feeling of the railway brake noise. In the proposed method, masker signal was designed by using the estimated noise at real time. Therefore, masker signal can follow the time variation of multiple spectral peaks. Moreover, the sound pressure level is determined by using SP_SNR to adaptively estimate the change of it. The time variation of masker signal is smoothed by referencing not only current frame but also several previous frames. For shown above reasons, we subjectively confirmed the effectiveness of the proposed method for non-stationary noise. In addition, in this experiment, the value of discomfort reduction which uses the masker signals with white noise and pink noise are evaluated. We confirmed sufficiently discomfort reduction using the masker signal with pink noise compared with the masker signal with white noise from a result of subjective evaluation experiment. It is assumed that pink noise has 1/f fluctuation which produces comfort feeling for human [10]. 6 Conclusions In this paper, we proposed a discomfort reduction mothod based on auditory masking with the adaptive noise estimation for the railway brake sound. We carried out subjective evaluation experiment with the actual railway brake sound. As a result, we confirmed the effectiveness of the proposed method. In future work, we have a plan to design masker signals by employing suitable original sound source of masker signal for sufficiently discomfort reduction. Moreover, we have a plan to improve the estimation accuracy of the adaptive noise estimation. 9

10 Acknowledgments The work was partly supported by R-GIRO (Ritsumeikan Global Innovation Research Organization) funded by Ritsumeikan University. References [1] Lorang, X; Foy-Margiocchi, F; Nguyen Q.S; Gautier, P.E, TGV disc brake squeal, Journal of Sound and Vibration, Vol. 293, No.3-5, [2] Desplanques, Y; Roussette, O; Degallaix, G; Copin, R; Berthier, Y, Analysis of tribological behaviour of pad disc contact in railway braking Part 1. Laboratory test development, compromises between actual and simulated tribological triplets, Wear, Vol. 262, No. 5, 2007, pp [3] Elliott, J; Nelson, A, Active noise control, Signal Processing Magazine, IEEE, Vol. 10, No. 4, [4] Tsujikawa, M; Morise, M; Nishiura, T, A study of comfortable sound design for narrow-band noise based on psychoacoustic evaluation criteria, Internoise2011, Osaka, Japan, September 4-7, 2011, Tue-P-21. [5] Suhara, Y; Ikefuji, D; Nakayama, M; Nishiura, T, A design of control signal in reducing discomfort of the dental treatment sound based on auditory masking, ICA2013, Montreal, Canada, June 2-7, 2013, PaperID:3pNSc8. [6] Wegel, R.L; Lane, E.C, The Auditory Masking of One Pure Tone by Another and its Probable Relation to the Dynamics of the Inner Ear, Physical Review, Vol. 23, 1924, pp [7] Kumer, S; Foster, H; Bailey, P; Griffiths, T, Mapping unpleasantness of sounds to their auditory representation, The Journal of the Acoustical Society of America, Vol. 124, No. 6, 2008, pp [8] Zwicker, E; Terhardt, E, Analytical expression for critical-band rate and critical band width as a function of frequency, The Journal of Acoustical Society of America, Vol. 68, No. 5, 1980, pp [9] Scheffe, H, An analysis of variance for paired comparisons, The Journal of the American Statistical Association, Vol. 47, No. 259, 1952, pp [10] Musha, T, 1/f Fluctuations, A Monthly Publication of the Japan Society of Applied Physics, Vol. 46, No. 12, 1977, pp