Voiceprint monitoring sensor optimization arrangement method suitable for small box girder modular expansion joint

Abstract

The invention discloses a voiceprint monitoring sensor optimization arrangement method suitable for a small box girder modular expansion joint. The optimized arrangement method of the voiceprint monitoring sensors comprises the following steps: preliminarily drawing up preset positions of the voiceprint monitoring sensors on and under a bridge; sound is collected through a voiceprint monitoring sensor, each section of sound comprises background noise and sound generated when a vehicle collides with the expansion joint, and a threshold value of the picked background noise is set; calculating an effective waveform energy ratio and a signal to interference plus noise ratio of the audio according to a set background noise threshold value and the audio waveform scatter points; and comparing the effective waveform energy ratio with the signal to interference plus noise ratio, and optimally arranging the sensor. According to the method, the arrangement of the voiceprint sensors is optimized, it is ensured that the collected vehicle impact sound signals are clear, stable and minimum in interference, meanwhile, the influence of parameter differences of equipment of multiple brands on the monitoring result is eliminated, the data consistency and reliability are improved, and therefore precise and intelligent diagnosis of expansion joint diseases is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of sensors, and particularly relates to a soundprint monitoring sensor optimal arrangement method suitable for small box girder modular expansion joints. BACKGROUND

[0002] As one of the important auxiliary facilities of bridge structures, expansion joints play a crucial role in the normal operation of bridges and the smooth passage of vehicles. However, due to the harsh service environment, their service life is often lower than expected, and they are extremely vulnerable components. The performance degradation of expansion joints directly affects the safety of the overall bridge structure and the smoothness of driving. Key diseases of expansion joints and their auxiliary structures include rubber support sliding, support beam and steel weld seam debonding, etc. If these diseases are not discovered in time, with the passage of time and the action of vehicle, environmental and other loads, the performance of the steel will gradually degrade, gradually reaching a critical dangerous state, and eventually leading to sudden fracture and buckling of the steel, causing a series of traffic accidents.

[0003] Currently, the damage detection of bridge expansion joints mainly relies on manual visual inspection. This traditional inspection method relies on the experience of the inspector and is highly subjective, and it also causes traffic interruption and affects the normal operation of the bridge. In addition, existing technical means have obvious shortcomings in the rapid diagnosis, efficient evaluation, economy and intelligence of the internal working state of the expansion joint, so there is an urgent need for a comprehensive diagnostic method that can realize real-time, accurate, low-cost and intelligent diagnosis. It is understood that experienced maintenance personnel mainly judge the working state of the expansion joint through hearing, so it is feasible to realize intelligent diagnosis of expansion joint diseases through effective analysis and data modeling of audio information.

[0004] Currently, the soundprint monitoring technology can realize intelligent diagnosis of the working state of the expansion joint, but the arrangement of the soundprint sensor in the existing technology lacks scientific basis and there is no clear arrangement method. How to find the soundprint sensor layout position with the clearest reflection, the most recognizable, the least interference, and the most stable characteristics of the impact sound of the vehicle passing through the expansion joint is a complex optimization problem that needs to consider multiple factors. The current arrangement of soundprint sensors leads to the following problems: the sensor layout position is highly random, making it difficult to ensure that the collected sound signals (vehicle impact expansion joint sound) have high recognizability and stability; background noise interference is large, and the noise reduction effect of sensors of different positions and brands differs significantly, making it difficult to uniformly evaluate the effectiveness of the data; there is a lack of quantitative indicators to optimize the sensor arrangement, and it is difficult to balance the monitoring accuracy and economy. SUMMARY

[0005] In view of the problems in the prior art, the present application provides an acoustic fingerprint monitoring sensor optimal arrangement method suitable for small box girder modular expansion joints, especially suitable for medium displacement modular expansion joints, which ensures that the collected vehicle impact sound signals are clear, stable and least disturbed by optimizing the arrangement position of the acoustic fingerprint monitoring sensor; and the advantages and disadvantages of different arrangement schemes are evaluated through quantitative indicators (such as effective waveform energy ratio and signal-to-interference noise ratio), so as to realize the optimization of the number and position of the sensor; the optimal arrangement method can eliminate the influence of parameter differences of multiple brands of equipment on the monitoring results, improve the consistency and reliability of the data, and thus realize accurate and intelligent diagnosis of the expansion joint disease.

[0006] In order to achieve the above-mentioned application purposes, the present application provides the following technical solutions:

[0007] An acoustic fingerprint monitoring sensor optimal arrangement method suitable for small box girder modular expansion joints, specifically comprising:

[0008] (1) Acoustic fingerprint monitoring sensor preset position: the acoustic fingerprint monitoring sensor (sound pickup) preset position is initially determined based on the on-site exploration results of the bridge, and the acoustic fingerprint monitoring sensor preset position includes the bridge and the bridge below, and in order to avoid the randomness of the results caused by the same type of sensor, at least two different brands of acoustic fingerprint monitoring sensors are arranged at the same position;

[0009] The bridge acoustic fingerprint monitoring sensor arrangement principle is to arrange the sensor in the area that does not hinder the normal traffic of vehicles and does not affect the normal work of the modular expansion joint, and the specific preset position is initially determined as the side of the lane guardrail;

[0010] The bridge acoustic fingerprint monitoring sensor arrangement principle is to arrange the sensor in the area that does not hinder the normal traffic of vehicles and does not affect the normal work of the modular expansion joint, and the specific preset position is initially determined as the side of the lane guardrail;

[0011] A camera is arranged beside each preset position to assist the audio cutting work and accurately intercept the time stamp of the vehicle passing through the expansion joint;

[0012] (2) Sound collection and analysis: collect sound through the acoustic fingerprint monitoring sensors preset in step (1), and each sound includes background noise and vehicle impact expansion joint sound; in view of the difference in noise reduction effect of different brands of acoustic fingerprint monitoring sensors at the same position and the difference in sound pickup effect of the same brand of acoustic fingerprint monitoring sensor at different positions, set a background noise threshold for each acoustic fingerprint monitoring sensor to distinguish effective sound and ineffective sound; the effective sound is the vehicle impact expansion joint sound, and the ineffective sound is the background noise;

[0013] The setting method of the background noise threshold adopts a sliding window dynamic percentile threshold method, which analyzes the amplitude distribution of the audio signal through a sliding window, dynamically calculates P95 (95th percentile) and the maximum value of each window, and then takes the median of the moving average smoothing as the final threshold; the core is to reflect the time-varying characteristics of the signal through local window statistics, combine the noise resistance of the percentile and the stability of the moving average, and realize the adaptive determination of the background noise threshold.

[0014] The specific setting method comprises the following steps:

[0015] ① Audio segmentation: each voiceprint monitoring sensor picks up an audio, which is automatically cut to ensure that the audio after cutting is a background noise segment; the background noise segment audio is divided into a plurality of audio short time periods, each audio short time period comprising a short time window and a frame shift; the number of audio short time periods is determined according to the length of the background noise segment audio, the length of the short time window and the frame shift;

[0016] The length of the short time window is 40ms-60ms;

[0017] The frame shift is 5ms-15ms;

[0018] ② Window internal statistical calculation: calculate P95 and the maximum value in each short time window to avoid the sensitivity of global statistics due to instantaneous changes;

[0019] P95 = Percentile (X window , 95%)

[0020] ③ Dynamic smoothing and threshold determination: apply a 3-point moving average filter to the obtained P95 sequence, wherein P 95k represents the 95th percentile of the kth short time window;

[0021]

[0022] Take the median of the smoothed P95 sequence as the final threshold;

[0023]

[0024] The advantages of the threshold setting method of the present application are:

[0025] ① Time domain adaptability: the background noise may change over time (such as environmental noise difference), and the sliding window can capture this dynamic characteristic;

[0026] ② Anti-instantaneous interference: the maximum value is easily affected by sudden noise, and P95 can effectively suppress abnormal points;

[0027] ③ Robustness of calculation: the median decision resists local fluctuations, while the mean method is sensitive to discrete points.

[0028] (3) Audio calculation: the audio time length picked up by each voiceprint monitoring sensor remains consistent, and the effective waveform energy ratio and signal-to-interference-plus-noise ratio (SINR) of each audio segment are calculated according to the set background noise threshold and audio waveform scatter points;

[0029] The audio time length is 0.7-0.8s;

[0030] The effective waveform energy ratio (EWER) refers to the ratio of the effective waveform area (total waveform area-background noise forming waveform area) to the total waveform area, and the value is between 0 and 1;

[0031]

[0032] The signal-to-interference-plus-noise ratio (SINR) refers to the ratio of the effective signal area to the background noise signal area, and the value is usually greater than 1;

[0033]

[0034] Where s(t) represents the waveform curve formed by the scatter points with audio waveform scatter points greater than the background noise threshold, and ∫ T |s(t)| 2 dt represents the area formed by the curve; similarly, n(t) represents the waveform curve formed by the scatter points with audio waveform scatter points less than the background noise threshold, and ∫ T |n(t)| 2 dt represents the area formed by the curve;

[0035] (4) Voiceprint monitoring sensor arrangement optimization: comparing the effective waveform energy ratio and signal-to-interference-plus-noise ratio calculated by the voiceprint monitoring sensors at different positions of the same brand, the larger the value, the more optimal the arrangement position of the voiceprint monitoring sensor of the brand.

[0036] Compared with the prior art, the present application has the following beneficial effects:

[0037] (1) The voiceprint monitoring sensor is optimally arranged by the method of the present application, ensuring that the vehicle impact sound signals collected by the sensor at the optimal arrangement position are clear, stable and least disturbed, while eliminating the influence of parameter differences of multiple brands on the monitoring results, improving the consistency and reliability of the data, and realizing accurate and intelligent diagnosis of the joint disease.

[0038] (2) The optimal arrangement method adopted by the present application is convenient and fast, and can be used as a reference for the arrangement of voiceprint monitoring sensors for other types of bridges (such as T-beams, continuous beams, etc.). BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 Technical roadmap of the voiceprint monitoring sensor arrangement method described in the present application.

[0040] Figure 2 Schematic diagram of the measurement point position of the voiceprint monitoring sensor described in the present application on a bridge.

[0041] Figure 3 Schematic diagram of the audio waveform picked up by the voiceprint monitoring sensor described in the present application.

[0042] Figure 4 Standardized waveform diagram of the audio picked up by different voiceprint monitoring sensors described in the present application, wherein the microphone numbers of each diagram are (a) fh_1, (b) fh_2, (c) fh_3, (d) fh_4, (e) hk_1, (f) hk_2, (g) hk_3, (h) hk_4, (i) yk_1, (j) yk_2, (k) yk_3, (l) net_sq1, (m) net_sq2, and (n) net_sq3.

[0043] Figure 5 Schematic diagram of the background noise threshold setting of the sensor number fh_4 described in the present application.

[0044] Figure 6 Schematic diagram of the background noise threshold setting of the sensor number hk_1 described in the present application.

[0045] Figure 7 Schematic diagram of the background noise threshold setting of the sensor number yk_1 described in the present application.

[0046] Figure 8 Schematic diagram of the background noise threshold setting of the sensor number net_sq2 described in the present application.

[0047] Figure 9 Schematic diagram of the sliding window when setting the background noise threshold described in the present application.

[0048] Figure 10 Voiceprint signal waveform diagram of the voiceprint monitoring sensor described in the present application.

[0049] Figure 11 Comparison of the measurement point position indicators of the brand 1 voiceprint monitoring sensor described in the present application.

[0050] Figure 12 Comparison of the measurement point position indicators of the brand 2 voiceprint monitoring sensor described in the present application.

[0051] Figure 13The brand 3 voiceprint monitoring sensor position index of each measuring point of the embodiment of the present application is compared.

[0052] Figure 14 The brand 4 voiceprint monitoring sensor position index of each measuring point of the embodiment of the present application is compared.

[0053] Wherein, Figure 2 Each mark in the figure is: 1 voiceprint monitoring sensor, 2 camera. DETAILED DESCRIPTION

[0054] The technical solutions of the present application are further described below in combination with specific embodiments and the accompanying drawings of the specification.

[0055] A voiceprint monitoring sensor optimization arrangement method suitable for small box girder modular expansion joints, specifically comprising:

[0056] (1) Voiceprint sensor preset position: the voiceprint monitoring sensor preset position is initially determined through the on-site exploration results of the bridge, the voiceprint monitoring sensor preset position includes the bridge and the bridge below, and in order to avoid the randomness of the results caused by the same type of sensor, at least two different brands of voiceprint monitoring sensors are arranged at the same position;

[0057] The bridge voiceprint monitoring sensor arrangement principle is: the sensor is arranged in the area which does not hinder the normal traffic of vehicles and does not affect the normal work of the modular expansion joint, and the specific preset position is initially determined as the side of the lane guardrail;

[0058] The bridge voiceprint monitoring sensor arrangement principle is: the sensor is arranged in the area which does not hinder the normal traffic of vehicles and does not affect the normal work of the modular expansion joint, and the specific preset position is initially determined as the side of the lane guardrail;

[0059] (2) Sound collection and analysis: the sound is collected by the voiceprint monitoring sensor initially determined at each position in step (1), and each sound includes background noise and vehicle impact expansion joint sound, and the background noise threshold value picked up by each voiceprint sensor is set to distinguish effective sound and ineffective sound according to the difference of the noise reduction effect of the voiceprint monitoring sensor of different brands at the same position and the difference of the sound pickup effect of the voiceprint monitoring sensor of the same brand at different positions; the effective sound is the vehicle impact expansion joint sound, and the ineffective sound is the background noise;

[0060] The background noise threshold is set by using a sliding window dynamic percentile thresholding method. This method analyzes the amplitude distribution of the audio signal through a sliding window, dynamically calculates the P95 (95th percentile) and maximum value of each window, and then uses the median as the final threshold after smoothing by moving average. Its core is to reflect the time-varying characteristics of the signal through local window statistics, and combine the noise resistance of percentiles and the stability of moving average to achieve adaptive determination of the background noise threshold.

[0061] The specific methods for setting the background noise threshold include:

[0062] Step 1 (Audio Segmentation): Divide the background noise segment audio (0.2s, sampling frequency 16000Hz) into short time windows (e.g., 50ms), frame shift (10ms), and divide it into 16 short audio segments in total;

[0063] Step 2 (Statistical Calculation within Window): Calculate P95 and maximum value within each window to avoid the sensitivity of global statistics to instantaneous changes;

[0064] P95 = Percentile(X) window Step 3 (Dynamic Smoothing and Threshold Determination): 95%

[0065] ① Apply a 3-point moving average filter to the obtained P95 sequence; where P 95k This represents the 95th percentile of the k-th window;

[0066]

[0067] ② Take the median of the smoothed P95 sequence as the final threshold;

[0068]

[0069] (3) Audio calculation: The audio pickup time of each voiceprint monitoring sensor is kept consistent. The effective waveform energy ratio and signal-to-interference-plus-noise ratio of each audio segment are calculated based on the set background noise threshold and audio waveform scatter points. The audio duration is 0.7 to 0.8 seconds.

[0070] The effective waveform energy ratio (EWER) refers to the ratio of the effective waveform area (total waveform area - waveform area formed by background noise) to the total waveform area, and its value is between 0 and 1.

[0071]

[0072] The signal-to-interference-plus-noise ratio (SINR) refers to the ratio of the effective signal area to the background noise signal area, which is usually greater than 1;

[0073]

[0074] Where s(t) represents the waveform curve formed by the scatter points of the audio waveform whose scatter points are greater than the background noise threshold, and ∫ T |s(t)| 2 dt represents the area formed by the curve; similarly, n(t) represents the waveform curve formed by the scatter points of the audio waveform whose scatter points are less than the background noise threshold, and ∫ T |n(t)| 2 dt represents the area formed by the curve;

[0075] (4) Voiceprint monitoring sensor arrangement optimization: compare the effective waveform energy ratio and signal-to-interference-plus-noise ratio calculated by the voiceprint monitoring sensors of the same brand in different positions. The larger the value, the more optimal the layout position of the voiceprint monitoring sensor of the brand.

[0076] Embodiment one

[0077] This embodiment is the arrangement of small box girder voiceprint monitoring sensors in a certain area of Zhenjiang. Specifically, a D160 type modular expansion joint of a certain bridge is selected as the test object, and the specific arrangement method includes:

[0078] (1) 3 measuring points under the bridge and 1 measuring point on the bridge. Each measuring point under the bridge is arranged with 4 products (brand 1, brand 2, brand 3, and brand 4), and there are a total of 12 sensors under the bridge. Each measuring point on the bridge is arranged with 2 products (brand 1 and brand 2), and there are a total of 2 sensors on the bridge. There are a total of 14 sensors on the bridge and under the bridge. Measuring point 1 under the bridge is the top surface of the small box girder concrete, measuring point 2 is the lower surface of the expansion joint steel, and measuring point 3 is the surface of the bridge bearing. The measuring point on the bridge is located beside the guardrail of the fast lane. The voiceprint monitoring sensor measuring point arrangement is shown in Figure 2 , and the layout position and corresponding number are shown in Table 1;

[0079] Table 1 Position of voiceprint monitoring sensor preset position and corresponding number

[0080]

[0081] (2) Threshold setting: The sound is collected by the voiceprint monitoring sensor preset at each position in step (1), and each piece of sound includes background noise and vehicle impact on the expansion joint sound. The threshold of the background noise collected by each voiceprint monitoring sensor is set to distinguish the effective sound and the invalid sound according to the difference of the noise reduction effect of the voiceprint monitoring sensor of different brands at the same position and the difference of the sound pickup effect of the voiceprint monitoring sensor of the same brand at different positions. The effective sound is the vehicle impact on the expansion joint sound, and the invalid sound is the background noise.

[0082] Each piece of audio is 0.75 s long. As shown in FIG. 2, the threshold is set by calculating the amplitude data distribution of the audio (background noise) in the first 0.2 s. In order to ensure that the first 0.2 s of the intercepted audio is a background noise segment, the sound collected by each voiceprint monitoring sensor is automatically cropped, and as shown in FIG. 3, 10 pieces of standardized audio are manually selected for background noise amplitude data distribution calculation. The background noise data distribution and threshold setting of different voiceprint monitoring sensors are shown in Table 2. Figure 3 Figure 4 Figures 5 to 9 The related schematic diagram for setting the background noise threshold is shown in FIG. 4.

[0083] Table 2 Background noise data distribution and threshold setting of different voiceprint sensors

[0084]

[0085]

[0086] (3) Audio calculation: In an ideal state, the sound collected by the voiceprint monitoring sensor includes background white noise and impact sound generated by the vehicle passing through the expansion joint. As shown in FIG. 5, the blue area is background noise, and the orange area is the effective sound of the vehicle passing through the expansion joint. The audio pickup time length of each voiceprint monitoring sensor remains consistent, and the effective waveform energy ratio and the signal-to-interference-plus-noise ratio of each piece of audio are calculated according to the set background noise threshold and the audio waveform scatter plot. Figure 10

[0087] The effective waveform energy ratio (EWER) refers to the ratio of the effective waveform area (total waveform area-background noise formed waveform area) to the total waveform area, and its value is between 0 and 1.

[0088]

[0089] The signal-to-interference-plus-noise ratio (SINR) refers to the ratio of the effective signal area to the background noise signal area, and its value is usually greater than 1. ​​​

[0090]

[0091] Where s(t) represents the waveform curve formed by scatter points of the audio waveform that are greater than the background noise threshold, ∫ T |s(t)| 2 dt represents the area formed by the curve; similarly, n(t) represents the waveform curve formed by scatter points whose points are smaller than the background noise threshold, ∫ T |n(t)| 2 dt represents the area formed by the curve;

[0092] (4) Optimization of the placement of voiceprint monitoring sensors: Compare the effective waveform energy ratio and signal-to-interference-plus-noise ratio of voiceprint monitoring sensors at different locations of the same brand. The larger the value, the better the placement of the voiceprint monitoring sensor of that brand.

[0093] like Figure 11 As shown, for Brand 1, the effective waveform energy ratio and signal-to-interference-plus-noise ratio of the sound collected by the soundprint monitoring sensors at the three locations under the bridge are significantly better than those on the bridge. The order of sound pickup effect at the three locations under the bridge is: bottom surface of the central beam steel structure > top surface of the pier cap > top surface of the small box girder concrete structure.

[0094] like Figure 12 As shown, for Brand 2, the effective waveform energy ratio and signal-to-interference-plus-noise ratio of the sound collected by the soundprint monitoring sensors at the three locations under the bridge are significantly better than those on the bridge. The order of sound pickup effect at the three locations under the bridge is: bottom surface of the central steel beam > top surface of the pier cap > top surface of the small box girder concrete.

[0095] like Figure 13 As shown, for brand 3, the sound pickup effect at the three locations under the bridge is in the following order: bottom surface of the central beam steel > top surface of the pier cap > top surface of the small box girder concrete. However, the effective waveform energy ratio and signal-to-interference-plus-noise ratio evaluation indicators are close to 0, indicating that the sound collected by this type of voiceprint monitoring sensor is basically background noise.

[0096] like Figure 14 As shown, for brand 4, the sound pickup effect at the three locations under the bridge is in the following order: bottom surface of the middle beam steel ≈ top surface of the pier cap > top surface of the small box girder concrete. The effective waveform energy ratio is close to 1 and the signal-to-interference-plus-noise ratio evaluation index is greater than 1.

Claims

1. A method for optimizing the arrangement of voiceprint monitoring sensors suitable for small box girder modular expansion joints, characterized by, Specifically comprising: (1) The preset position of the acoustic fingerprint monitoring sensor: the preset position of the acoustic fingerprint monitoring sensor is initially determined based on the field exploration results of the bridge, and the preset position of the acoustic fingerprint monitoring sensor includes the bridge and the bridge below, and at least two acoustic fingerprint monitoring sensors of different brands are arranged at the same position; The principle of arranging the acoustic fingerprint monitoring sensor on the bridge is to arrange the sensor in the area where the vehicle can pass normally and the modular expansion joint can work normally; The principle of arranging the acoustic fingerprint monitoring sensor under the bridge is to arrange the sensor in the area where the vehicle can pass normally and the modular expansion joint can work normally, and the energy attenuation of the sound generated by the vehicle hitting the expansion joint is the smallest in the propagation process, the signal distortion degree is low, and the original acoustic fingerprint characteristics can be restored to the maximum extent; (2) Sound collection and analysis: collect the sound through the acoustic fingerprint monitoring sensor preset in step (1), and each sound includes background noise and vehicle impact expansion joint sound, and set the background noise threshold value picked up by each acoustic fingerprint monitoring sensor to distinguish effective sound and ineffective sound; The effective sound is the vehicle impact expansion joint sound, and the ineffective sound is the background noise; The setting method of the background noise threshold value adopts a sliding window dynamic percentile threshold method, which analyzes the amplitude distribution of the audio signal through a sliding window, dynamically calculates the P95 and the maximum value of each window, and then takes the median of the smoothed P95 sequence as the final threshold value; (3) Audio calculation: the audio length picked up by each acoustic fingerprint monitoring sensor is consistent, and the effective waveform energy ratio and the signal-to-noise ratio of each audio are calculated according to the set background noise threshold value and the audio waveform scatter; The effective waveform energy ratio refers to the ratio of the effective waveform area to the total waveform area, and the value is between 0 and 1; The signal-to-noise ratio refers to the ratio of the effective signal area to the background noise signal area, and the value is usually greater than 1; where s(t) represents the waveform curve formed by the points of the audio waveform that are greater than the background noise threshold, and T |s(t)| 2 dt represents the area under the waveform curve; similarly, n(t) represents the waveform curve formed by the points of the audio waveform that are less than the background noise threshold, and T |n(t)| 2 dt represents the area under the waveform curve; (4) Acoustic fingerprint monitoring sensor arrangement optimization: compare the effective waveform energy ratio and the signal-to-noise ratio calculated by the acoustic fingerprint monitoring sensors of the same brand at different positions, and the larger the value, the better the position for the acoustic fingerprint monitoring sensor of the brand.

2. The voiceprint monitoring sensor optimization arrangement method of claim 1, wherein, The specific preset position of the acoustic fingerprint monitoring sensor on the bridge in step (1) is initially determined as the side of the lane guardrail; The specific preset position of the acoustic fingerprint monitoring sensor under the bridge is initially determined as the bottom surface of the expansion joint type steel, the bottom surface of the small box girder concrete, and the surface of the bridge bearing platform.

3. The voiceprint monitoring sensor optimization arrangement method of claim 1, wherein, A camera is arranged beside each preset position of the acoustic fingerprint monitoring sensor in step (1).

4. The voiceprint monitoring sensor optimization arrangement method of claim 1, wherein, The setting method of the background noise threshold value in step (2) specifically includes the following steps: ① Audio segmentation: cut the audio picked up by each acoustic fingerprint monitoring sensor to ensure that the audio after cutting is a background noise segment; The background noise segment audio is divided into a plurality of audio short periods, and each audio short period includes a short time window and a frame shift; the number of audio short periods is determined according to the length of the background noise segment audio, the length of the short time window, and the frame shift; The length of the short time window is 40ms-60ms; The frame shift is 5ms-15ms; ② Window statistics calculation: calculate the P95 and the maximum value in each short time window; P95 = Percentile (X window , 95%) iii. Dynamic smoothing and threshold determination: Apply a 3-point moving average filter to the resulting P95 sequence, where P 95k denotes the 95th percentile of the kth short-time window. Take the median of the smoothed P95 sequence as the final threshold value; 5. The voiceprint monitoring sensor optimization arrangement method of claim 1, wherein, The audio duration in step (3) is 0.7-0.8s.

Images