Intelligent earphone adaptive performance detection method and system based on data analysis

Abstract

The present application relates to the technical field of intelligent earphone performance detection, and specifically relates to an intelligent earphone adaptive performance detection method and system based on data analysis, which comprises obtaining a wearer speaking sound segment, a response sound segment, a media reference sound segment, and a response interval duration; wearing a to-be-tested head-mounted intelligent earphone on a head acoustic simulation device and inputting the media reference sound segment; applying a wearer speaking incentive and a dialogue party response incentive, collecting an in-ear response signal to obtain an in-ear response sequence; determining a media attenuation completion time according to the media reference sound segment and the in-ear response sequence, and determining a response transmission establishment completion time according to the response sound segment and the in-ear response sequence; determining a later time as a dialogue acceptance formation time, and generating an adaptive dialogue acceptance performance detection result. The present application can quantify a dialogue acceptance process after a self-speaking trigger, and improve the accuracy of intelligent earphone adaptive performance detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent headphone performance testing technology, specifically to an adaptive performance testing method and system for intelligent headphones based on data analysis. Background Technology

[0002] With the development of interactive functions in smart headphones, some smart headphones now possess adaptive conversation listening capabilities triggered by the wearer's own speech. When the wearer speaks, the headphones automatically switch from sound-isolated listening mode to conversation listening mode, reducing the impact of continuous sound isolation on face-to-face conversations. Traditional headphone performance testing primarily takes place in an anechoic chamber or test box, using an artificial ear to measure the static acoustic parameters of the headphones under preset states such as passive sound isolation, active noise cancellation, and transparency mode. This method assumes that the headphone's operating mode has been manually set and does not involve the adaptive transition process between different operating states.

[0003] However, in real-world interactive scenarios where the wearer's voice triggers the headphones to enter a conversational listening state, there is a continuous temporal relationship between the wearer's voice trigger, the headphones' listening state transition, and the arrival of the other party's response. If it is only confirmed whether the headphones have switched modes, it is difficult to determine whether the conversational listening state formed after the headphones have completed the adaptive transition can promptly receive the subsequent response from the other party.

[0004] Therefore, there is an urgent need for a data analysis-based method and system for detecting the adaptive performance of smart headphones, which can objectively detect the adaptive dialogue continuity performance of smart headphones in scenarios where the wearer initiates a conversation. Summary of the Invention

[0005] The purpose of this invention is to provide a data analysis-based method and system for detecting the adaptive performance of smart headphones, so as to achieve objective quantitative detection of whether the headphones promptly receive the responding voice of the other party in the dialogue listening state when the wearer's self-speech triggers the headphones' adaptive switching.

[0006] To achieve the above objectives, in one aspect, the present invention provides a method for detecting the adaptive performance of smart headphones based on data analysis, the method comprising: Step S1: Obtain the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; wear the smart headphones under test on the head acoustic simulation device, control the smart headphones under test to be in a soundproof listening state and enable the self-speaking trigger dialogue listening adaptive function, and input the media reference voice segment into the smart headphones under test.

[0007] Step S2: Apply the wearer's speaking stimulus corresponding to the wearer's speaking voice segment to the simulated mouth position of the head acoustic simulation device. After the response interval, apply the dialogue party's response stimulus corresponding to the response voice segment in front of the head acoustic simulation device. Record the wearer's speaking start time, wearer's speaking end time, response start time, and response end time. Collect the in-ear response signal output by the head acoustic simulation device to obtain the in-ear response sequence.

[0008] Step S3: Construct a media residual curve based on the media reference sound segment and the in-ear response sequence. Determine the media attenuation completion time based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking. Construct a response input curve based on the response sound segment and the in-ear response sequence. Determine the response input establishment completion time based on the rising segment and high-level stable segment of the response input curve after the response start time. Determine the later of the media attenuation completion time and the response input establishment completion time as the reception formation time. Generate an adaptive dialogue reception performance detection result based on the reception formation time, response start time, and response end time, including the duration of the first unreceived segment of the response and the duration of the effective reception of the response.

[0009] Furthermore, the method of applying the wearer's speaking stimulus corresponding to the wearer's speaking voice segment to the simulated mouth position of the head acoustic simulation device, and applying the dialogue party's response stimulus corresponding to the response voice segment in front of the head acoustic simulation device after a response interval, and recording the wearer's speaking start time, wearer's speaking end time, response start time, and response end time includes: The low-frequency band of the wearer's speech segment is extracted to obtain the low-frequency speech component; the structural vibration driving signal is generated based on the low-frequency speech component and the vibration amplitude coefficient; the wearer's speech segment and the structural vibration driving signal are synchronized at the same application start time to obtain composite speech excitation control data.

[0010] Based on the composite speech excitation control data, oral acoustic excitation corresponding to the wearer's speech segment is applied at the simulated mouth position of the head acoustic simulation device, and head vibration excitation corresponding to the structural vibration drive signal is applied at the jaw structure position of the head acoustic simulation device; the moment when the oral acoustic excitation begins to be applied is recorded as the wearer's speech start time, and the moment when the oral acoustic excitation ends to be applied is recorded as the wearer's speech end time.

[0011] After the response interval has elapsed since the wearer finished speaking, the corresponding response stimulus of the dialogue party is applied in front of the head acoustic simulation device; the moment when the dialogue party response stimulus is applied is recorded as the response start time, and the moment when the dialogue party response stimulus is applied is recorded as the response end time.

[0012] Furthermore, the method for determining the vibration amplitude coefficient includes: Obtain the candidate coefficient sequence of vibration amplitude; generate candidate structure vibration driving signals based on the low-frequency speech component and each candidate coefficient of vibration amplitude in the candidate coefficient sequence; control the smart headphones under test to be in soundproof listening state and enable the self-speech triggered dialogue listening adaptive function, and input the media reference sound segment to the smart headphones under test.

[0013] According to the preset repetition cycle, the wearer's speaking voice segment and the vibration drive signal of each candidate structure are synchronously applied to the head acoustic simulation device, and the calibration intra-ear response sequence output by the head acoustic simulation device is collected; a calibration media residual curve is constructed based on the media reference voice segment and each calibration intra-ear response sequence; the media attenuation start time sequence corresponding to each vibration amplitude candidate coefficient in the preset repetition cycle is determined based on the continuous descent start point of the calibration media residual curve, and vibration amplitude candidate coefficients that do not form a media attenuation start time sequence are eliminated.

[0014] The trigger dispersion is calculated based on the media attenuation start time sequence. Among the vibration amplitude candidate coefficients in the media attenuation start time sequence where the media attenuation start time is within the preset trigger time range, the vibration amplitude candidate coefficient with the smallest trigger dispersion is selected. Among the vibration amplitude candidate coefficients with the same trigger dispersion, the vibration amplitude candidate coefficient with the smallest value is selected as the vibration amplitude coefficient.

[0015] Furthermore, the method for constructing a media residual curve based on a media reference sound segment and an in-ear response sequence, and determining the media attenuation completion time based on the falling segment and low-level stable segment of the media residual curve during its change process after the wearer begins speaking, includes: The wearer's speech start time is determined based on the composite speech excitation control data; according to the sampling time synchronization relationship between the media reference sound segment and the in-ear response sequence, the media steady-state response segment is extracted from the in-ear response sequence before the wearer's speech start time, and the media reference segment corresponding to the media steady-state response segment is extracted from the media reference sound segment.

[0016] The media playback time shift and frequency division amplitude compensation vector are calculated based on the media reference segment and the media steady-state response segment; the media reference segment is time-aligned and frequency band amplitude is corrected based on the media playback time shift and frequency division amplitude compensation vector to obtain the corrected media reference segment; the media attenuation analysis sequence after the wearer's speech start moment is extracted from the in-ear response sequence.

[0017] Within a continuous analysis window, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is calculated to obtain a media matching intensity sequence arranged according to sampling time. The media matching intensity sequence is then determined as the media residual curve. The media attenuation completion time is determined based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking.

[0018] Furthermore, the method for determining the completion time of media attenuation based on the descending segment and the low-level stable segment of the media residual curve after the wearer begins speaking includes: The media residual curve is differentially calculated according to the preset first analysis step size to obtain a coarse-grained media change rate sequence; the media decrease candidate interval is determined according to the time intervals in the coarse-grained media change rate sequence that are continuously less than the preset decrease change rate threshold.

[0019] Within the media descent candidate interval and the extended analysis interval following the media descent candidate interval, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is recalculated according to a preset second analysis step size to obtain the local media residual curve. The preset second analysis step size is smaller than the preset first analysis step size.

[0020] The local media change rate sequence and local media fluctuation amplitude sequence are calculated based on the local media residual curve. The moment when the local media change rate sequence changes from being continuously less than the preset decreasing change rate threshold to being continuously less than the preset stable change rate threshold in absolute value, and the local media fluctuation amplitude sequence is continuously less than the preset stable fluctuation threshold, is determined as the media attenuation completion moment.

[0021] Furthermore, the method for constructing a response input curve based on the response sound segment and the in-ear response sequence, and determining the completion time of response input establishment based on the growth segment and high-level stable segment of the response input curve after the response initiation time, includes: The response segment is windowed according to the preset analysis window length to obtain the response reference window sequence; the reference effective frequency band energy of each response reference window in the response reference window sequence is calculated to obtain the reference effective frequency band energy sequence; response reference windows with reference effective frequency band energy greater than the response analysis energy threshold are selected from the response reference window sequence, and the time position of each response reference window in the response segment is retained to obtain the effective response reference window sequence.

[0022] Based on the time position of the sound segment corresponding to the effective response reference window sequence, extract the response input candidate response sequence after the response start time from the intra-ear response sequence; construct the response input curve based on the effective response reference window sequence and the response input candidate response sequence.

[0023] The completion time of response establishment is determined by the growth segment and the high-level stable segment of the response input curve after the response start time.

[0024] Furthermore, the method for constructing the response input curve based on the effective response reference window sequence and the response input candidate response sequence includes: Based on the sound segment time position and preset time delay search range of each effective response reference window, the candidate response window group corresponding to each effective response reference window is extracted from the response input candidate response sequence; the normalized correlation coefficient between each effective response reference window and each candidate response window in the corresponding candidate response window group is calculated to obtain the candidate matching strength group corresponding to each effective response reference window.

[0025] The maximum candidate matching strength in each candidate matching strength group is determined as the response input matching strength of the corresponding valid response reference window, and the center time of the candidate response window corresponding to the maximum candidate matching strength is determined as the input response time of the corresponding valid response reference window; the response input matching strengths are arranged according to each input response time to obtain the response input curve.

[0026] The response input change rate is calculated based on the difference in matching strength between adjacent response inputs and the time difference between adjacent input response times, resulting in a response input change rate sequence. The response input change rate sequence is transformed from continuously exceeding a preset growth change rate threshold to continuously being less than a preset stable input change rate threshold in absolute value, and the fluctuation amplitude of the response input curve in the corresponding time interval is continuously less than a preset stable input fluctuation threshold. This transformation time is defined as the response input establishment completion time.

[0027] Based on the same inventive concept, this invention also provides a data analysis-based intelligent headphone adaptive performance detection system, the system comprising: The test configuration module is used to acquire the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; the smart headphones under test are worn on the head acoustic simulation device, the smart headphones under test are controlled to be in a soundproof listening state and the self-speaking triggered dialogue listening adaptive function is enabled, and the media reference voice segment is input to the smart headphones under test.

[0028] The interactive stimulus and response acquisition module is used to apply the wearer's speech stimulus corresponding to the wearer's speech segment to the simulated mouth position of the head acoustic simulation device, and after the response interval, apply the dialogue party's response stimulus corresponding to the response segment in front of the head acoustic simulation device, record the wearer's speech start time, wearer's speech end time, response start time, and response end time; and acquire the in-ear response signal output by the head acoustic simulation device to obtain the in-ear response sequence.

[0029] The hyperbolic reception analysis module is used to construct a media residual curve based on the media reference sound segment and the in-ear response sequence. It determines the media attenuation completion time based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking. It also constructs a response input curve based on the response sound segment and the in-ear response sequence. The response input establishment completion time is determined based on the rising segment and high-level stable segment of the response input curve after the response start time. The later of the media attenuation completion time and the response input establishment completion time is determined as the reception formation time. Finally, it generates an adaptive dialogue reception performance test result based on the reception formation time, response start time, and response end time, including the duration of the first unreceived segment of the response and the effective reception duration of the response.

[0030] Compared with the prior art, the beneficial effects of the present invention are: 1. By constructing a detection process for the connection between the wearer's speech trigger and the responder's response, the completion time of media attenuation and the completion time of response input establishment are determined respectively. The later time is taken as the connection formation time, thereby generating the duration of the first segment of the response that is not connected and the duration of the effective connection of the response. This can quantify the state transition process of the smart earphone after the speech trigger and the actual connection relationship of the responder's voice, avoiding the evaluation of adaptive performance based solely on whether the mode is switched, and improving the accuracy of the detection of the adaptive dialogue connection performance of smart earphones in face-to-face dialogue scenarios.

[0031] 2. By performing time-shift correction and frequency division amplitude correction on the media reference sound segment through the media steady-state response segment before the wearer begins speaking, a media residue curve that is more consistent with the actual media residue in the ear is constructed. Combined with coarse-grained candidate interval localization and local fine analysis, the media attenuation completion time is determined. This can reduce the impact of playback link delay, headphone frequency response differences and high-resolution calculation throughout the entire time period on media attenuation identification, and improve the localization accuracy and calculation efficiency of the media attenuation completion time. Attached Figure Description

[0032] Figure 1 This is a flowchart of the data analysis-based adaptive performance testing method for smart headphones according to the present invention; Figure 2 This is a block diagram of the data analysis-based intelligent headphone adaptive performance testing system of the present invention. Figure 3 This is a schematic diagram of the hyperbolic continuation analysis process of the intraocular response sequence of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Before giving examples, it is necessary to explain the application scenarios of the present invention. The present invention is applicable to the performance testing scenarios of smart headphones with self-speech triggering dialogue listening adaptive function, especially applicable to the testing scenarios where the smart headphones under test are in media playback and soundproof listening state, and after the wearer speaks, the smart headphones under test automatically switch to dialogue listening state, and it is necessary to evaluate whether the subsequent response of the dialogue party is received in a timely manner.

[0035] Example 1: As Figure 1 As shown in the figure, this embodiment provides a data analysis-based method for detecting the adaptive performance of smart headphones. The method includes: Acquire the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; place the smart headphones under test on the head acoustic simulation device, control the smart headphones under test to be in a soundproof listening state and enable the self-speaking trigger dialogue listening adaptive function, and input the media reference voice segment into the smart headphones under test.

[0036] The smart headset under test, designated H1, features active noise cancellation, transparent listening, and self-adaptive dialogue response triggered by speech. The test enclosure houses a head acoustic simulation device, a simulated mouth test sound source, a dialogue response test sound source, and a signal analyzer. The head acoustic simulation device includes a left artificial ear, a right artificial ear, and simulated mouth and jaw positions. The left and right artificial ears are connected to two acoustic acquisition channels of the signal analyzer, with a sampling frequency of 48kHz for each channel. The left earcup of the smart headset H1 covers the left artificial ear, and the right earcup covers the right artificial ear. The wearing position of the smart headset H1 relative to the head acoustic simulation device remains constant throughout the testing process.

[0037] Both the wearer's speaking and response segments were extracted from natural speech recordings at a sampling frequency of 48kHz. The wearer's speaking segment consisted of a 1.20s natural speech phrase, "What did you just say?", while the response segment consisted of a 1.60s natural speech phrase, "I said the meeting will begin in ten minutes." The media reference segment consisted of an 8.00s broadband music clip, sampled at 48kHz, maintaining a continuous energy distribution within the 300Hz to 6000Hz frequency band. The response interval was set to 0.35s, starting at the end of the wearer's speaking segment.

[0038] The tested H1 smart headphones establish a Bluetooth connection with the audio input terminal. The accompanying test software sends a media reference audio segment to the audio input terminal, which then sends the media reference audio segment back to the H1 smart headphones. The playback volume of the H1 smart headphones is set to 60% of the device's nominal maximum volume. Active noise cancellation is enabled, transparency listening is disabled, and self-talk-triggered dialogue listening adaptive function is enabled. The H1 smart headphones are in a soundproof listening state, formed by the superposition of media playback and active noise cancellation. After the media reference audio segment begins input, the accompanying test software maintains a 2.00s stable media playback period. During this 2.00s stable media playback period, the simulated mouth test sound source and the dialogue party's response test sound source are not driven. The duration of the media reference audio segment is greater than the sum of the stable media playback period, the wearer's speech segment, the response interval, and the duration of the response segment.

[0039] The low-frequency band of the wearer's speech is extracted to obtain the low-frequency speech component; a structural vibration driving signal is generated based on the low-frequency speech component and the vibration amplitude coefficient; the method for determining the vibration amplitude coefficient includes: Obtain the candidate coefficient sequence of vibration amplitude; generate candidate structure vibration driving signals based on the low-frequency speech component and each candidate coefficient of vibration amplitude in the candidate coefficient sequence; control the smart headphones under test to be in soundproof listening state and enable the self-speech triggered dialogue listening adaptive function, and input the media reference sound segment to the smart headphones under test.

[0040] According to the preset repetition cycle, the wearer's speaking voice segment and the vibration drive signal of each candidate structure are synchronously applied to the head acoustic simulation device, and the calibration intra-ear response sequence output by the head acoustic simulation device is collected; a calibration media residual curve is constructed based on the media reference voice segment and each calibration intra-ear response sequence; the media attenuation start time sequence corresponding to each vibration amplitude candidate coefficient in the preset repetition cycle is determined based on the continuous descent start point of the calibration media residual curve, and vibration amplitude candidate coefficients that do not form a media attenuation start time sequence are eliminated.

[0041] The trigger dispersion is calculated based on the media attenuation start time sequence. Among the vibration amplitude candidate coefficients in the media attenuation start time sequence where all media attenuation start times fall within the preset trigger time range, the vibration amplitude candidate coefficient with the smallest trigger dispersion is selected. Among the vibration amplitude candidate coefficients with the same trigger dispersion, the vibration amplitude candidate coefficient with the smallest value is selected as the vibration amplitude coefficient. The wearer's speaking voice segment and the structural vibration drive signal are synchronized according to the same application start time to obtain composite speaking excitation control data.

[0042] Based on the composite speech excitation control data, oral acoustic excitation corresponding to the wearer's speech segment is applied at the simulated mouth position of the head acoustic simulation device, and head vibration excitation corresponding to the structural vibration drive signal is applied at the jaw structure position of the head acoustic simulation device; the moment when the oral acoustic excitation begins to be applied is recorded as the wearer's speech start time, and the moment when the oral acoustic excitation ends to be applied is recorded as the wearer's speech end time.

[0043] After the response interval has elapsed since the wearer finished speaking, a response stimulus corresponding to the response segment is applied in front of the head acoustic simulation device. The moment the response stimulus begins to be applied is recorded as the response start time, and the moment the response stimulus ends is recorded as the response end time. The intra-auricular response signal output by the head acoustic simulation device is collected to obtain the intra-auricular response sequence.

[0044] The wearer's voice is processed by a fourth-order Butterworth bandpass filter, with the lower bandpass cutoff frequency set at 80Hz and the upper bandpass cutoff frequency set at 320Hz, yielding the low-frequency speech component. This low-frequency speech component retains the low-frequency fluctuations in the wearer's voice that are synchronized with changes in vocal vibration. The discrete sample values ​​of the low-frequency speech component are... The absolute value of the peak value of the low-frequency speech component is denoted as Normalized low-frequency speech components The reference driving voltage of the vibration drive channel is set to 1.00V, the vibration amplitude coefficient is k, and the discrete sampling value of the structural vibration drive signal is... A linear vibration exciter is fixed in position on the jaw structure. The drive input terminal of the linear vibration exciter is connected to the vibration drive channel controlled by the supporting test software. The structural vibration drive signal is sent to the linear vibration exciter through the vibration drive channel.

[0045] The candidate coefficient sequence for vibration amplitude is set to {0.20, 0.35, 0.50, 0.65, 0.80}. Each candidate coefficient for vibration amplitude is substituted into... The vibration driving signal of the candidate structure was calculated. The preset repetition rounds were set to 5 rounds. Each candidate coefficient of vibration amplitude corresponds to 5 calibration triggering processes. Each calibration triggering process uses the tested smart headphones H1, the media reference sound segment, the wearer's speaking sound segment, and a 2.00s stable media playback period. The media reference sound segment is replayed at the beginning of each calibration triggering process. The oral acoustic excitation is applied by a simulated oral test sound source, and the head vibration excitation is applied by a linear vibration exciter. The oral acoustic excitation and head vibration excitation are started synchronously using the same digital trigger edge. The A-weighted equivalent sound level of the oral acoustic excitation is calibrated to 68dB at 0.20m in front of the simulated oral position.

[0046] The calibration sampling signal output from the left artificial ear is denoted as the calibration intra-auricular response sequence. The media reference segment is divided into a media calibration reference window using an 80ms analysis window length and a 10ms analysis step size. The calibration intra-auricular response sequence is divided into media calibration response windows using the same analysis window length and analysis step size. The normalized correlation coefficient between the media calibration reference window and the media calibration response window is... The calculation formula is: .

[0047] in, For the first The sampled values ​​of each media calibration reference window, For the first The sampled values ​​of each media calibration response window, For the first The average value of each media calibration reference window. For the first The average value of each media calibration response window. The number of sampling points within the 80ms analysis window was 3840. The normalized correlation coefficients were arranged according to sampling time to obtain the calibration media residual curve.

[0048] The starting point for the continuous decline is taken as the center moment of the first analysis window on the residual curve of the calibration medium that satisfies the condition of three consecutive declines. The condition of three consecutive declines is set as follows: , , The media attenuation start time is obtained by subtracting the moment when the acoustic excitation at the mouth is first applied from the center time of the analysis window corresponding to the sustained descent start point. In the five rounds of calibration media residual curves corresponding to the vibration amplitude candidate coefficient 0.20, three rounds did not form a sustained descent start point; therefore, the vibration amplitude candidate coefficient 0.20 was discarded. The vibration amplitude candidate coefficients 0.35, 0.50, 0.65, and 0.80 all formed media attenuation start time sequences, which are as follows: ; ; ; .

[0049] Trigger dispersion is represented by the sample standard deviation of the media decay initiation time sequence. Vibration amplitude candidate coefficients. Corresponding trigger dispersion ;in, , Candidate coefficients for vibration amplitude In the The start time of media attenuation during the wheel calibration trigger process. This represents the average value of the media attenuation start time sequence. The trigger dispersions corresponding to the vibration amplitude candidate coefficients of 0.35, 0.50, 0.65, and 0.80 are 0.043s, 0.016s, 0.011s, and 0.007s, respectively. The preset trigger time range is set to 0.35s to 0.70s. The media attenuation start time sequence corresponding to the vibration amplitude candidate coefficient of 0.35 contains 0.74s, 0.79s, and 0.76s, which exceed the preset trigger time range; the media attenuation start times corresponding to the vibration amplitude candidate coefficients of 0.50, 0.65, and 0.80 are all within the preset trigger time range. The vibration amplitude candidate coefficient of 0.80 has the smallest trigger dispersion, and the vibration amplitude coefficient is determined to be 0.80.

[0050] The structural vibration driving signal is The wearer's speech segment and the structural vibration drive signal share the same application start time. The accompanying test software sends the wearer's speech segment to the simulated mouth test sound source and the structural vibration drive signal to the vibration drive channel to obtain composite speech excitation control data. The simulated mouth test sound source is located at the simulated mouth position of the head acoustic simulation device, and the linear vibration exciter is located at the jaw structure position of the head acoustic simulation device. The accompanying test software outputs a synchronous trigger command at 2.00s when the media reference sound segment is played, and the moment when the mouth acoustic excitation begins to be applied is recorded as the wearer's speech start time. The oral acoustic excitation lasted for 1.20 seconds, and the moment the oral acoustic excitation ended was recorded as the moment the wearer finished speaking. .

[0051] The sound source for the dialogue response test was positioned 1.20m directly in front of the head acoustic simulation device, with the center height of the sound source aligned with the center height of the simulated mouth. The A-weighted equivalent sound level of the dialogue response excitation was calibrated to 60dB at the center of the line connecting the two ears of the head acoustic simulation device. The response interval was 0.35s, and the moment the dialogue response excitation began to be applied was recorded as the response start time. The response segment lasted 1.60 seconds, and the moment when the dialogue party stopped applying the response stimulus was recorded as the response end time. The signal analyzer synchronously acquires the output signals of the left and right artificial ears from the start of the media reference sound segment input, with the acquisition ending 1.00 seconds after the response end time. The left artificial ear output signal is recorded as the intraocular response sequence, with a total recording duration of 6.15 seconds. The right artificial ear output signal is acquired using the same acquisition rules to obtain the right ear intraocular response sequence. The intraocular response sequence covers the media reference sound segment input, the application of the wearer's speaking stimulus, the application of the dialogue party's response stimulus, and the response process after the dialogue party's response stimulus ends. Figure 3 As shown, both the media residual curve and the response input curve are extracted from the intra-auricular response sequence. The media residual curve is constructed by the matching relationship between the corrected media reference sound segment and the intra-auricular response sequence, and the response input curve is constructed by the matching relationship between the response sound segment and the intra-auricular response sequence. The media residual curve is used to determine the media attenuation completion time, and the response input curve is used to determine the response input establishment completion time. The later of the media attenuation completion time and the response input establishment completion time is determined as the reception formation time.

[0052] The wearer's speech start time is determined based on the composite speech excitation control data; according to the sampling time synchronization relationship between the media reference sound segment and the in-ear response sequence, the media steady-state response segment is extracted from the in-ear response sequence before the wearer's speech start time, and the media reference segment corresponding to the media steady-state response segment is extracted from the media reference sound segment.

[0053] The media playback time shift and frequency division amplitude compensation vector are calculated based on the media reference segment and the media steady-state response segment; the media reference segment is time-aligned and frequency band amplitude is corrected based on the media playback time shift and frequency division amplitude compensation vector to obtain the corrected media reference segment; the media attenuation analysis sequence after the wearer's speech start moment is extracted from the in-ear response sequence.

[0054] Within a continuous analysis window, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is calculated to obtain a media matching intensity sequence arranged according to sampling time. The media residual curve is differentially calculated according to a preset first analysis step size to obtain a coarse-grained media change rate sequence. The media descent candidate interval is determined based on the time intervals in the coarse-grained media change rate sequence that are continuously less than a preset descent rate threshold.

[0055] Within the media descent candidate interval and the extended analysis interval following the media descent candidate interval, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is recalculated according to a preset second analysis step size to obtain the local media residual curve. The preset second analysis step size is smaller than the preset first analysis step size.

[0056] The local media change rate sequence and local media fluctuation amplitude sequence are calculated based on the local media residual curve. The moment when the local media change rate sequence changes from being continuously less than the preset decreasing change rate threshold to being continuously less than the preset stable change rate threshold in absolute value, and the local media fluctuation amplitude sequence is continuously less than the preset stable fluctuation threshold, is determined as the media attenuation completion moment.

[0057] The synchronous application start time for the composite speech stimulus control data recording is 2.00s, and the start time of the wearer's speech is determined to be... The media reference segment is input starting at 0.00s, and the stable playback range is from 0.50s to 1.70s after the input of the media reference segment. The sampled data from 0.50s to 1.70s in the in-ear response sequence is extracted as the media steady-state response segment, and the sampled data from 0.50s to 1.70s in the media reference segment is extracted as the media reference segment. The duration of both the media steady-state response segment and the media reference segment is 1.20s.

[0058] Media reference clips are The media steady-state response segment is The sampling frequency is 48kHz. Cross-correlation calculations are performed on the media reference segment and the media steady-state response segment within a time-shift search range of 0ms to 300ms. The cross-correlation function is: ;in, The number of sampling points for media reference segments. The number of candidate time-shift sampling points is defined as the number of candidate time-shift sampling points at which the cross-correlation function reaches its maximum value. The cross-correlation function reaches its maximum value when the number of candidate time-shift sampling points is 7488. 7488 sampling points correspond to 156ms, and the media playback time-shift is determined to be 0.156s. The media playback time-shift corresponds to the overall time shift of the media reference audio segment through the audio input terminal, Bluetooth transmission link, playback link of the tested smart headphones H1, and the left ear implant acquisition link.

[0059] After the media reference segment and the media steady-state response segment are aligned for 0.156 seconds, the frequency division amplitude compensation vector is calculated according to six frequency division intervals within the 300Hz to 6000Hz frequency band. The six frequency division intervals are set as follows: 300Hz to 600Hz, 600Hz to 1000Hz, 1000Hz to 1600Hz, 1600Hz to 2500Hz, 2500Hz to 4000Hz, and 4000Hz to 6000Hz. The root mean square amplitude of the media reference segment within each frequency division interval is , No. The root mean square amplitude of the media steady-state response segment within each frequency division interval is , No. The frequency division amplitude compensation coefficient corresponding to each frequency division interval is: The frequency division amplitude compensation vector calculated from the six frequency division intervals is G={0.86,0.93,1.04,1.01,0.94,0.88}.

[0060] The media reference segment is shifted backward by 0.156s based on the media playback time shift. Then, amplitude correction is applied to each of the six frequency division intervals according to the frequency division amplitude compensation vector. The signals from the six frequency division intervals after amplitude correction are summed to obtain the corrected media reference segment. The corrected media reference segment and the in-ear response sequence are on the same sampling time axis. The sampling data from 2.00s to 5.15s in the in-ear response sequence is truncated into a media attenuation analysis sequence. The media attenuation analysis sequence covers the wearer's speech start time, wearer's speech end time, response start time, and response end time. The analysis window length of the continuous analysis window is set to 120ms, and the preset first analysis step size is set to 40ms. The corrected media reference segment extracts the media reference analysis window according to the analysis window length of 120ms and the preset first analysis step size of 40ms. The media attenuation analysis sequence extracts the media response analysis window according to the same analysis window length and the same analysis step size. The i-th media reference analysis window and the i-th... Normalized correlation coefficients among media response analysis windows The calculation formula is: .

[0061] in, For the first Sampled values ​​from each media reference analysis window. For the first Sampled values ​​from each media response analysis window. For the first The average value of each media reference analysis window. For the first The average value of each media response analysis window. The number of sampling points corresponding to the 120ms analysis window is 5760. Normalized correlation coefficient. Arranged according to the center time of the analysis window, the media matching intensity sequence is obtained, and the media matching intensity sequence is denoted as the media residual curve.

[0062] The media residual curves at 2.34s, 2.38s, 2.42s, 2.46s, 2.50s, 2.54s, 2.58s, 2.62s, 2.66s, 2.70s, 2.74s, 2.78s, and 2.82s correspond to media matching intensities of 0.918, 0.906, 0.854, 0.742, 0.615, 0.487, 0.355, 0.258, 0.201, 0.169, 0.153, 0.145, and 0.142, respectively. The coarse-grained media change rate sequence is... The preset threshold for the rate of decrease is set to... The coarse-grained media change rate sequence after the wearer begins speaking is searched for time intervals where the rate of change is continuously less than a preset decreasing rate of change threshold. The time interval with the longest continuous duration is identified as the media decreasing candidate interval. The coarse-grained media change rate sequence is consistently less than the preset decreasing rate of change threshold within the continuous time interval of 2.38s to 2.66s, and this 2.38s to 2.66s interval is the longest continuous time interval; therefore, the media decreasing candidate interval is determined to be 2.38s to 2.66s. The extended analysis interval is set to 0.24s after the end of the media decreasing candidate interval, and is therefore determined to be 2.66s to 2.90s.

[0063] The media descent candidate interval and the extended analysis interval are merged to obtain a local fine analysis interval from 2.38s to 2.90s. The preset second analysis step size is set to 10ms. The corrected media reference segment and the media attenuation analysis sequence continue to use an analysis window of 120ms within the local fine analysis interval, and the normalized correlation coefficient is recalculated according to the preset second analysis step size of 10ms to obtain the local media residual curve. The media matching intensities corresponding to the local media residual curve at 2.70s, 2.71s, 2.72s, 2.73s, 2.74s, 2.75s, 2.76s, 2.77s, 2.78s, and 2.79s are 0.207, 0.194, 0.181, 0.168, 0.155, 0.143, 0.142, 0.141, 0.141, and 0.140, respectively.

[0064] The local media change rate sequence is , The first in the local media residual curve The local media match strength is calculated based on the difference between the maximum and minimum values ​​of five consecutive media match strengths. Local media fluctuations The preset stable rate of change threshold is set to... The preset stable fluctuation threshold is set to 0.010. The local media change rate sequence is continuously less than the preset decreasing change rate threshold within the corresponding interval from 2.70s to 2.74s. The local media change rate sequence continuously satisfies the condition that the absolute value is less than the preset stable change rate threshold from 2.75s onwards. The local media fluctuation amplitude sequence is continuously less than the preset stable fluctuation threshold from 2.75s onwards. The media decay completion time is determined as... .

[0065] The response segment is windowed according to the preset analysis window length to obtain the response reference window sequence; the reference effective frequency band energy of each response reference window in the response reference window sequence is calculated to obtain the reference effective frequency band energy sequence; response reference windows with reference effective frequency band energy greater than the response analysis energy threshold are selected from the response reference window sequence, and the time position of each response reference window in the response segment is retained to obtain the effective response reference window sequence.

[0066] Based on the time position of the sound segment corresponding to the effective response reference window sequence, the response input candidate response sequence after the response start time is extracted from the in-ear response sequence; based on the time position of the sound segment of each effective response reference window and the preset time delay search range, the candidate response window group corresponding to each effective response reference window is extracted from the response input candidate response sequence; the normalized correlation coefficient between each effective response reference window and each candidate response window in the corresponding candidate response window group is calculated respectively to obtain the candidate matching strength group corresponding to each effective response reference window.

[0067] The maximum candidate matching strength in each candidate matching strength group is determined as the response input matching strength of the corresponding valid response reference window, and the center time of the candidate response window corresponding to the maximum candidate matching strength is determined as the input response time of the corresponding valid response reference window; the response input matching strengths are arranged according to each input response time to obtain the response input curve.

[0068] The response input change rate is calculated based on the difference in matching strength between adjacent responses and the time difference between adjacent responses, resulting in a response input change rate sequence. The response input change rate sequence is then transformed from continuously exceeding a preset growth rate threshold to continuously being less than a preset stable input rate threshold, and the fluctuation amplitude of the response input curve within the corresponding time interval is continuously less than a preset stable input fluctuation threshold. This transformation is defined as the response input establishment completion time. The response input establishment completion time is determined based on the growth segment and high-level stable segment of the response input curve after the response start time. The later of the media attenuation completion time and the response input establishment completion time is defined as the reception formation time. An adaptive dialogue reception performance detection result is generated based on the reception formation time, response start time, and response end time, including the duration of the first unreceived segment of the response and the duration of the effective reception of the response.

[0069] The response segment is divided into a response reference window sequence with a preset analysis window length of 120ms and a windowing step size of 40ms. The sampling frequency of the response segment is 48kHz, and 120ms corresponds to 5760 sampling points. Each response reference window is multiplied by a Hanning window and then subjected to a Fast Fourier Transform to calculate the spectral energy in the 300Hz to 4000Hz frequency band. Effective bandwidth energy of each response reference window ;in, For the first Each response reference window at frequency Spectral amplitude at that location This indicates the discrete frequency points of the Fast Fourier Transform with a center frequency in the 300Hz to 4000Hz frequency band.

[0070] The reference effective bandwidth energy is expressed in normalized form, the first... Reference effective frequency band energy of each response reference window ;in, The total number of response reference windows in the response reference window sequence is 38. The response segment lasts for 1.60 s, and 38 response reference windows are obtained according to the preset analysis window length of 120 ms and the windowing step size of 40 ms. The response analysis energy threshold is set to 0.20. The reference effective frequency band energies corresponding to the response reference windows located at the beginning of the response segment at 0.06 s, 0.10 s, 0.14 s, 0.18 s, 0.22 s, 0.26 s, 0.30 s, 0.34 s, 0.38 s, and 0.42 s are 0.14, 0.23, 0.36, 0.49, 0.62, 0.74, 0.81, 0.77, 0.69, and 0.58, respectively. The response reference window located at 0.06s from the start of the response segment was removed because the reference effective frequency band energy was less than the response analysis energy threshold. Response reference windows located at 0.10s, 0.14s, 0.18s, 0.22s, 0.26s, 0.30s, 0.34s, 0.38s, and 0.42s from the start of the response segment were retained. Of the 38 response reference windows, 31 met the condition that the reference effective frequency band energy was greater than the response analysis energy threshold. These 31 response reference windows, along with their segment time positions within the response segment, constitute the effective response reference window sequence.

[0071] The response start time is The response end time is The preset time delay search range is set to 0.00s to 0.25s. The candidate response sequence is sampled from 3.55s to 5.40s in the in-ear response sequence. 5.40s is obtained by adding the response end time of 5.15s to the upper limit of the preset time delay search range of 0.25s. Each valid response reference window maintains a window length of 120ms. The effective response reference window with a sound segment time position of u_b, within the candidate search center time range of the candidate response sequence, is represented as follows: Candidate response window groups are extracted sequentially within the candidate search center time range with a search step size of 10ms, and the window length of the candidate response window group is also set to 120ms.

[0072] No. One valid response reference window is , No. The first valid response reference window corresponds to the first The candidate response window is . No. The first effective response reference window and the first Normalized correlation coefficient among candidate response windows ;in, The average value of the b-th valid response reference window. For the first The average value of each candidate response window. All normalized correlation coefficients corresponding to the first effective response reference window are arranged in the search order of the candidate response windows to obtain the first... A candidate matching strength group.

[0073] The nominal arrival time corresponding to the effective response reference window with a sound segment time position of 0.22s is The candidate response window group covers the period from 3.77s to 4.02s. The maximum normalized correlation coefficient in the candidate response window group is 0.63, and the center time of the candidate response window corresponding to the maximum value is 3.78s. 0.63 is determined as the response input matching intensity corresponding to the effective response reference window at the sound segment time position of 0.22s. The response input matching intensity is extracted for each effective response reference window according to the same rules, and the center time of the candidate response window corresponding to the maximum candidate matching intensity is recorded. The center time of the candidate response window corresponding to the maximum candidate matching intensity is used as the input response time of the corresponding effective response reference window, and the response input matching intensity and the input response time are combined to form the response input curve points. The response input curve points are arranged in ascending order of input response time to obtain the response input curve.

[0074] The response-input matching intensities at 3.66s, 3.70s, 3.74s, 3.78s, 3.82s, 3.86s, 3.90s, 3.94s, 3.98s, and 4.02s are 0.20, 0.32, 0.48, 0.63, 0.76, 0.84, 0.848, 0.851, 0.850, and 0.852, respectively. The time difference between the input and response times corresponding to adjacent response-input matching intensities is 0.040s. The response-input change rate sequence... ;in, For the first Each response input matches the strength. This represents the time difference between the incoming response times corresponding to the matching strength of adjacent responses. The rate of change of incoming responses in the interval from 3.66s to 3.86s is as follows: , , , and The response input change rates corresponding to the interval from 3.86s to 4.02s are as follows: , , and The preset growth rate threshold is set to... The preset input stable rate of change threshold is set to .

[0075] The fluctuation amplitude of the response input curve is represented by the difference between the maximum and minimum values ​​among four consecutive response input matching intensities. The response input matching intensities corresponding to 3.90s, 3.94s, 3.98s, and 4.02s are grouped together, with an input fluctuation amplitude of 0.852 - 0.848 = 0.004; the preset input stability fluctuation threshold is set to 0.010. The response input change rate sequence is continuously greater than the preset growth change rate threshold within the interval from 3.66s to 3.86s; the response input change rate sequence continuously satisfies the condition that the absolute value is less than the preset input stability change rate threshold from 3.90s onwards; and the fluctuation amplitude formed by the four consecutive response input matching intensities corresponding to 3.90s onwards is less than the preset input stability fluctuation threshold. The time when the response input establishment is completed is determined as follows. .

[0076] The media decay completion time is The response input completion time is Then the formation time is The duration of the unresolved first paragraph of the response is... The effective response time is .Will , and Substitute and get , The adaptive dialogue acceptance performance test results were recorded as follows: the duration of the first segment of the response not being accepted was 0.35s, and the duration of the response being effectively accepted was 1.25s.

[0077] Example 2: Based on the same inventive concept, such as Figure 2 As shown, this embodiment also provides a smart headphone adaptive performance detection system based on data analysis, the system comprising: The test configuration module is used to acquire the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; the smart headphones under test are worn on the head acoustic simulation device, the smart headphones under test are controlled to be in a soundproof listening state and the self-speaking triggered dialogue listening adaptive function is enabled, and the media reference voice segment is input to the smart headphones under test.

[0078] The interactive stimulus and response acquisition module is used to apply the wearer's speech stimulus corresponding to the wearer's speech segment to the simulated mouth position of the head acoustic simulation device, and after the response interval, apply the dialogue party's response stimulus corresponding to the response segment in front of the head acoustic simulation device, record the wearer's speech start time, wearer's speech end time, response start time, and response end time; and acquire the in-ear response signal output by the head acoustic simulation device to obtain the in-ear response sequence.

[0079] The hyperbolic reception analysis module is used to construct a media residual curve based on the media reference sound segment and the in-ear response sequence. It determines the media attenuation completion time based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking. It also constructs a response input curve based on the response sound segment and the in-ear response sequence. The response input establishment completion time is determined based on the rising segment and high-level stable segment of the response input curve after the response start time. The later of the media attenuation completion time and the response input establishment completion time is determined as the reception formation time. Finally, it generates an adaptive dialogue reception performance test result based on the reception formation time, response start time, and response end time, including the duration of the first unreceived segment of the response and the effective reception duration of the response.

[0080] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0081] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data analysis-based method for detecting the adaptive performance of smart headphones, characterized in that, The method includes: Acquire the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; place the smart headphones under test on the head acoustic simulation device, control the smart headphones under test to be in a soundproof listening state and enable the self-speaking trigger dialogue listening adaptive function, and input the media reference voice segment into the smart headphones under test. A speech stimulus corresponding to the wearer's speech segment is applied to the simulated mouth position of the head acoustic simulation device. After the response interval, a response stimulus corresponding to the dialogue party's response segment is applied in front of the head acoustic simulation device. The start time of the wearer's speech, the end time of the wearer's speech, the start time of the response, and the end time of the response are recorded. The in-ear response signal output by the head acoustic simulation device is collected to obtain the in-ear response sequence. A media persistence curve is constructed based on the media reference sound segment and the in-ear response sequence. The media attenuation completion time is determined based on the falling segment and low-level stable segment of the media persistence curve after the wearer begins speaking. A response input curve is constructed based on the response sound segment and the in-ear response sequence. The response input establishment completion time is determined based on the rising segment and high-level stable segment of the response input curve after the response start time. The later of the media attenuation completion time and the response input establishment completion time is determined as the reception formation time. An adaptive dialogue reception performance test result is generated based on the reception formation time, the response start time, and the response end time, including the duration of the first unreceived segment of the response and the duration of the effective reception of the response.

2. The method for detecting the adaptive performance of smart headphones based on data analysis according to claim 1, characterized in that, The method of applying a wearer's speaking stimulus corresponding to the wearer's speaking voice segment to the simulated mouth position of the head acoustic simulation device, and applying a response stimulus corresponding to the dialogue party's response voice segment to the front of the head acoustic simulation device after a response interval, and recording the wearer's speaking start time, wearer's speaking end time, response start time, and response end time includes: Low-frequency bands are extracted from the wearer's speech segment to obtain low-frequency speech components; structural vibration driving signals are generated based on low-frequency speech components and vibration amplitude coefficients; the wearer's speech segment and structural vibration driving signals are synchronously configured at the same application start time to obtain composite speech excitation control data. Based on the composite speech excitation control data, oral acoustic excitation corresponding to the wearer's speech segment is applied to the simulated mouth position of the head acoustic simulation device, and head vibration excitation corresponding to the structural vibration drive signal is applied to the jaw structure position of the head acoustic simulation device; the moment when the oral acoustic excitation begins to be applied is recorded as the wearer's speech start time, and the moment when the oral acoustic excitation ends to be applied is recorded as the wearer's speech end time. After the response interval has elapsed since the wearer finished speaking, the corresponding response stimulus of the dialogue party is applied in front of the head acoustic simulation device; the moment when the dialogue party response stimulus is applied is recorded as the response start time, and the moment when the dialogue party response stimulus is applied is recorded as the response end time.

3. The method for detecting the adaptive performance of smart headphones based on data analysis according to claim 2, characterized in that, The method for determining the vibration amplitude coefficient includes: Obtain the vibration amplitude candidate coefficient sequence; generate candidate structure vibration driving signals based on the low-frequency speech component and each vibration amplitude candidate coefficient in the vibration amplitude candidate coefficient sequence; control the smart headphones under test to be in soundproof listening state and enable the self-speech triggered dialogue listening adaptive function, and input the media reference sound segment to the smart headphones under test; According to the preset repetition cycle, the wearer's voice segment and the vibration drive signal of each candidate structure are synchronously applied to the head acoustic simulation device, and the calibration intra-ear response sequence output by the head acoustic simulation device is collected; a calibration media residual curve is constructed based on the media reference sound segment and each calibration intra-ear response sequence; the media attenuation start time sequence corresponding to each vibration amplitude candidate coefficient in the preset repetition cycle is determined based on the continuous descent start point of the calibration media residual curve, and vibration amplitude candidate coefficients that do not form a media attenuation start time sequence are eliminated; The trigger dispersion is calculated based on the media attenuation start time sequence. Among the vibration amplitude candidate coefficients in the media attenuation start time sequence where the media attenuation start time is within the preset trigger time range, the vibration amplitude candidate coefficient with the smallest trigger dispersion is selected. Among the vibration amplitude candidate coefficients with the same trigger dispersion, the vibration amplitude candidate coefficient with the smallest value is selected as the vibration amplitude coefficient.

4. The data analysis-based adaptive performance testing method for smart headphones according to claim 3, characterized in that, The method for constructing a media persistence curve based on a media reference sound segment and an in-ear response sequence, and determining the media attenuation completion time based on the falling segment and low-level stable segment of the media persistence curve during the change process after the wearer begins speaking, includes: The wearer's speech start time is determined based on the composite speech excitation control data; according to the sampling time synchronization relationship between the media reference sound segment and the in-ear response sequence, the media steady-state response segment is extracted from the in-ear response sequence before the wearer's speech start time, and the media reference segment corresponding to the media steady-state response segment is extracted from the media reference sound segment; Calculate the media playback time shift and frequency division amplitude compensation vector based on the media reference segment and the media steady-state response segment; perform time alignment and frequency band amplitude correction on the media reference segment based on the media playback time shift and frequency division amplitude compensation vector to obtain the corrected media reference segment; extract the media attenuation analysis sequence after the wearer's speech start moment from the in-ear response sequence; Within a continuous analysis window, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is calculated to obtain a media matching intensity sequence arranged according to sampling time. The media matching intensity sequence is then determined as the media residual curve. The media attenuation completion time is determined based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking.

5. The method for detecting the adaptive performance of smart headphones based on data analysis according to claim 4, characterized in that, The method for determining the completion time of media attenuation based on the descending segment and the low-level stable segment of the media residual curve after the wearer begins speaking includes: The media residual curve is differentially calculated according to the preset first analysis step size to obtain a coarse-grained media change rate sequence; the media decrease candidate interval is determined according to the time intervals in the coarse-grained media change rate sequence that are continuously less than the preset decrease change rate threshold. Within the media descent candidate interval and the extended analysis interval following the media descent candidate interval, the normalized correlation coefficient between the corrected media reference segment and the media attenuation analysis sequence is recalculated according to the preset second analysis step size to obtain the local media residual curve. The preset second analysis step size is smaller than the preset first analysis step size. The local media change rate sequence and local media fluctuation amplitude sequence are calculated based on the local media residual curve. The moment when the local media change rate sequence changes from being continuously less than the preset decreasing change rate threshold to being continuously less than the preset stable change rate threshold in absolute value, and the local media fluctuation amplitude sequence is continuously less than the preset stable fluctuation threshold, is determined as the media attenuation completion moment.

6. The method for detecting the adaptive performance of smart headphones based on data analysis according to claim 5, characterized in that, The method for constructing a response aspiration curve based on the response sound segment and the in-ear response sequence, and determining the completion time of response aspiration establishment based on the growth segment and high-level stable segment of the response aspiration curve after the response initiation time, includes: The response segment is windowed according to the preset analysis window length to obtain the response reference window sequence; the reference effective frequency band energy of each response reference window in the response reference window sequence is calculated to obtain the reference effective frequency band energy sequence; response reference windows with reference effective frequency band energy greater than the response analysis energy threshold are selected from the response reference window sequence, and the time position of each response reference window in the response segment is retained to obtain the effective response reference window sequence. Based on the time position of the sound segment corresponding to the effective response reference window sequence, extract the response input candidate response sequence after the response start time from the intra-ear response sequence; construct the response input curve based on the effective response reference window sequence and the response input candidate response sequence. The completion time of response establishment is determined by the growth segment and the high-level stable segment of the response input curve after the response start time.

7. The method for detecting the adaptive performance of smart headphones based on data analysis according to claim 6, characterized in that, The method for constructing a response ingress curve based on the effective response reference window sequence and the response ingress candidate response sequence includes: Based on the sound segment time position and preset time delay search range of each effective response reference window, the candidate response window group corresponding to each effective response reference window is extracted from the response input candidate response sequence; the normalized correlation coefficient between each effective response reference window and each candidate response window in the corresponding candidate response window group is calculated respectively to obtain the candidate matching strength group corresponding to each effective response reference window. The maximum candidate matching strength in each candidate matching strength group is determined as the response input matching strength of the corresponding valid response reference window, and the center time of the candidate response window corresponding to the maximum candidate matching strength is determined as the input response time of the corresponding valid response reference window; the response input matching strengths are arranged according to each input response time to obtain the response input curve; The response input change rate is calculated based on the difference in matching strength between adjacent response inputs and the time difference between adjacent input response times, resulting in a response input change rate sequence. The response input change rate sequence is transformed from continuously exceeding a preset growth change rate threshold to continuously being less than a preset stable input change rate threshold in absolute value, and the fluctuation amplitude of the response input curve in the corresponding time interval is continuously less than a preset stable input fluctuation threshold. This transformation time is defined as the response input establishment completion time.

8. A data analysis-based intelligent headphone adaptive performance testing system, characterized in that, The system includes: The test configuration module is used to acquire the wearer's speaking voice segment, response voice segment, media reference voice segment, and response interval duration; the smart headphones under test are worn on the head acoustic simulation device, the smart headphones under test are controlled to be in a soundproof listening state and the self-speaking trigger dialogue listening adaptive function is enabled, and the media reference voice segment is input to the smart headphones under test; The interactive stimulus and response acquisition module is used to apply the wearer's speech stimulus corresponding to the wearer's speech segment to the simulated mouth position of the head acoustic simulation device, and apply the dialogue party's response stimulus corresponding to the response segment in front of the head acoustic simulation device after the response interval. It records the wearer's speech start time, wearer's speech end time, response start time, and response end time; and acquires the in-ear response signal output by the head acoustic simulation device to obtain the in-ear response sequence. The hyperbolic reception analysis module is used to construct a media residual curve based on the media reference sound segment and the in-ear response sequence. It determines the media attenuation completion time based on the falling segment and low-level stable segment of the media residual curve after the wearer begins speaking. It also constructs a response input curve based on the response sound segment and the in-ear response sequence. The response input establishment completion time is determined based on the rising segment and high-level stable segment of the response input curve after the response start time. The later of the media attenuation completion time and the response input establishment completion time is determined as the reception formation time. Finally, it generates an adaptive dialogue reception performance test result based on the reception formation time, response start time, and response end time, including the duration of the first unreceived segment of the response and the effective reception duration of the response.

Images