Earphone defect detection method and device based on voiceprint feature analysis and storage medium

Abstract

The invention discloses an earphone defect detection method and device based on voiceprint feature analysis and a storage medium, and relates to the technical field of electroacoustic quality control. According to the method, human ear hearing characteristics are simulated, Mel-cepstrum voiceprint characteristics are extracted from sound response signals, an adaptive weight mechanism based on an equal-loudness curve is introduced, and a voiceprint abnormal deviation index model conforming to human ear subjective perception is constructed; in order to solve the problem that tiny abnormal noise is difficult to recognize, high-order statistics such as kurtosis and skewness are adopted for feature fusion enhancement, and a support vector machine is combined to achieve accurate capture of transient impulse noise. According to the invention, various defect types such as voice coil friction, imbalance of balance degree, diaphragm damage and the like can be fully automatically distinguished, the problems that a traditional test means is disjointed with subjective hearing sense and the omission ratio of hidden defects is high are effectively solved, and the quality inspection efficiency and accuracy of earphone production lines are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of electroacoustic quality control, and in particular to a method, apparatus and storage medium for detecting headphone defects based on voiceprint feature analysis. Background Technology

[0002] As consumers' demands for headphone sound quality continue to rise, defect detection on headphone production lines has become increasingly important. Currently, headphone testing mainly relies on manual listening or simple frequency response curve comparison. Manual listening is greatly affected by subjective factors and fatigue, easily leading to misjudgments and missed detections, and is also inefficient. While traditional electroacoustic testing instruments can detect whether the frequency response curve is within tolerance, they are often insensitive to minute pure-tone abnormalities caused by voice coil friction or short-term pulse transient noise, because these defects may be masked by the main signal in the energy spectrum. In addition, existing automated testing methods lack simulation of human hearing characteristics, resulting in frequent "false negatives" where the instrument determines the product is qualified but the human ear perceives it poorly. This makes it difficult to accurately distinguish different types of acoustic defects and meet the high-precision quality inspection requirements of high-end headphone production lines. Summary of the Invention

[0003] To address the aforementioned problems in existing technologies, the present invention aims to provide a headphone defect detection method based on voiceprint feature analysis, the method comprising the following steps: Step S1: Drive the earphone under test to play a standard sweep frequency signal in an anechoic test environment, and use an artificial ear to collect the acoustic response signal of the earphone under test.

[0004] Step S2: Perform time-domain endpoint detection and frequency-domain noise reduction on the acoustic response signal to generate an audio frame sequence to be analyzed.

[0005] Step S3: Map the audio frame sequence to the Mel frequency domain and extract the voiceprint feature vector of each frame to construct the voiceprint feature matrix of the headphone to be detected.

[0006] Step S4: Obtain the reference voiceprint model of the standard headphones. Based on the voiceprint feature matrix and the reference voiceprint model, calculate the voiceprint anomaly deviation index of the headphones to be tested using a weighted distance algorithm.

[0007] Step S5: Compare the voiceprint abnormality deviation index with the preset defect classification threshold, and determine the defect type and quality level of the earphone to be tested based on the comparison result.

[0008] Preferably, step S3 includes the following sub-steps: Step S301: Perform a fast Fourier transform on each audio frame in the audio frame sequence to calculate the power spectral density function.

[0009] Step S302: Filter the power spectral density function through a preset Mel filter bank to obtain the log-Mel spectral energy value.

[0010] Step S303: Perform a discrete cosine transform on the logarithmic Mel spectrum energy value and retain the low-order coefficients as the voiceprint feature vector.

[0011] Preferably, in step S4, the formula for calculating the voiceprint anomaly deviation index is: ,in, This represents the voiceprint anomaly deviation index; This represents the total number of dimensions of the voiceprint feature vector; Indicates that the earphone to be tested is in the first... Feature coefficient values ​​in each feature dimension; Indicates that the standard headphones are in the first Reference feature coefficient values ​​in each feature dimension; Indicates the first The importance weight coefficients of each feature dimension.

[0012] The calculation formula quantifies the degree of sound quality deviation of the headphone under test by accumulating the weighted squared differences of each feature dimension.

[0013] Preferably, step S1 includes configuring the frequency range of the standard sweep signal to cover 20 Hz to 2 Hz and setting the logarithmic sweep mode; synchronously recording environmental background noise data during the acquisition process; and using an adaptive filter to filter out the environmental background noise data from the acoustic response signal to eliminate interference from the test environment on the voiceprint extraction.

[0014] Preferably, step S2 includes: calculating the short-time energy and short-time zero-crossing rate of the acoustic response signal; identifying the start and end points of the effective speech segments of the acoustic response signal based on the dual-threshold decision method of the short-time energy and the short-time zero-crossing rate; pre-emphasizing the extracted effective speech segments and performing frame-by-frame windowing operation using the Hamming window function to obtain the audio frame sequence.

[0015] Preferably, the importance weighting coefficient The setting logic is as follows: Obtain equal loudness curve data of human hearing; according to the first... The actual physical frequencies corresponding to each feature dimension are used to find the corresponding auditory sensitivity values ​​in the equal loudness curve data; these auditory sensitivity values ​​are then normalized and assigned to the importance weight coefficients. This makes the characteristic differences in the frequency bands sensitive to human ears contribute more to the voiceprint abnormality deviation index.

[0016] Preferably, step S5 includes: establishing a multi-level threshold table containing a pure tone abnormality threshold, a balance deviation threshold, and a distortion threshold; if the voiceprint abnormality deviation index exceeds the pure tone abnormality threshold, the defect type is determined to be voice coil friction or diaphragm damage; if the voiceprint abnormality deviation index is less than the pure tone abnormality threshold but exceeds the balance deviation threshold, the defect type is determined to be left and right ear sensitivity mismatch.

[0017] Preferably, the method further includes an enhanced detection step for transient abnormal sounds: in step S3, higher-order statistical features of the acoustic response signal are extracted, the higher-order statistical features including at least kurtosis coefficient and skewness coefficient; the higher-order statistical features are fused with the voiceprint feature vector; and a support vector machine classifier is used to identify the fused features to detect minute abnormal sound defects of short-time pulse type.

[0018] A headphone defect detection device based on voiceprint feature analysis, the device comprising: The acoustic data acquisition module is used to control the sound source to play a standard sweep frequency signal in an anechoic test environment and simultaneously acquire the acoustic response signal of the headphone under test.

[0019] The signal optimization processing module is used to perform time-domain endpoint detection and frequency-domain noise reduction processing on the acoustic response signal to generate a clean audio frame sequence.

[0020] The voiceprint matrix construction module is used to map the audio frame sequence to the Mel frequency domain and extract features to construct the voiceprint feature matrix of the headphone to be detected.

[0021] The anomaly index calculation module is used to call a pre-stored standard headphone reference voiceprint model to calculate the voiceprint anomaly deviation index of the headphone to be tested.

[0022] The defect classification and determination module is used to compare the voiceprint abnormality deviation index with the preset defect classification threshold to output the defect type and quality level of the earphone to be tested.

[0023] A computer-readable storage medium storing computer program instructions thereon, characterized in that the computer program instructions are read and executed by one or more processors, causing the one or more processors to perform the steps of a headphone defect detection method based on voiceprint feature analysis.

[0024] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a voiceprint analysis framework based on the characteristics of human auditory perception. By introducing a Mel filter bank to extract voiceprint features, and using equal loudness curve data to dynamically adjust the importance weight coefficients of each feature dimension, the calculation of the voiceprint abnormality deviation index is more consistent with the sensitivity of the human ear to different frequency bands. This weighting mechanism effectively solves the problem of inconsistency between traditional frequency response testing and subjective listening perception, and significantly improves the detection rate of minor defects that affect listening quality.

[0025] This invention proposes a multi-dimensional defect grading and enhanced detection mechanism. It not only achieves automated classification of common defects such as pure tone abnormalities and balance deviations through a multi-level threshold table, but also innovatively introduces high-order statistical features such as kurtosis and skewness to fuse with voiceprint features. Through endpoint detection, adaptive filtering and noise reduction, and feature fusion, the system can accurately capture highly concealed short-time pulse defects such as voice coil rubbing and diaphragm damage in noisy industrial environments, realizing a leap from single-index testing to comprehensive acoustic fingerprint diagnosis. Attached Figure Description

[0026] Figure 1 These are exemplary steps of the defect detection method of the present invention.

[0027] Figure 2 This is an exemplary flowchart of the steps for constructing the voiceprint feature matrix according to the present invention.

[0028] Figure 3 This is a schematic diagram of the module configuration of the detection device of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to specific embodiments.

[0030] like Figure 1 As shown, this invention provides a headphone defect detection method based on voiceprint feature analysis. The method includes the following steps: Step S1: Drive the earphone under test to play a standard sweep frequency signal in an anechoic test environment, and use an artificial ear to collect the acoustic response signal of the earphone under test. In an optional embodiment, the anechoic test environment is an anechoic box or a fully anechoic chamber with background noise below 15 dB, and its inner wall is lined with wedge-shaped sound-absorbing foam to isolate external electromagnetic interference and environmental noise, ensuring the signal-to-noise ratio of the test. The earphone under test is worn on an artificial ear simulator that conforms to the IEC-60318-4 or IEC-711 international standards, and the artificial ear integrates a high-sensitivity, wide-frequency response measurement microphone.

[0031] Step S1 includes configuring a standard sweep signal with a frequency range covering 20 Hz to 2 Hz and setting a logarithmic scan mode; synchronously recording ambient background noise data during the acquisition process; and using an adaptive filter to filter out the ambient background noise data from the acoustic response signal to eliminate interference from the test environment on the sound signature extraction. For example, the standard sweep signal is set as a continuous sine wave smoothly transitioning from a low frequency of 20 Hz to a high frequency of 20 kHz, with a duration set to 2 to 5 seconds to ensure sufficient vibration of the speaker unit. The logarithmic scan mode allows the low-frequency band, such as 20 Hz to 500 Hz, to occupy more time slices, thereby better covering the low-frequency details perceived by the human ear. The adaptive filter can employ a normalized least mean square algorithm, using a reference microphone placed outside the test chamber to pick up ambient noise and cancel out fan noise or motor vibration inside the test chamber in real time.

[0032] Step S2 involves performing time-domain endpoint detection and frequency-domain noise reduction on the acoustic response signal to generate an audio frame sequence to be analyzed. In practice, the acoustic response signal is typically a digital audio stream in WAV or PCM format with a quantization precision of 24 bits. It is recommended to set the sampling rate to 48 kHz or 96 kHz to ensure that high-frequency harmonic distortion information is acquired.

[0033] Step S2 includes calculating the short-time energy and short-time zero-crossing rate of the acoustic response signal; identifying the start and end points of the effective speech segments of the acoustic response signal based on the dual-threshold decision method of short-time energy and short-time zero-crossing rate; pre-emphasizing the extracted effective speech segments; and performing frame-by-frame windowing operation using the Hamming window function to obtain the audio frame sequence.

[0034] For example, the dual-threshold decision method sets two energy thresholds, high and low. First, the high threshold is used to roughly locate the signal range, and then the low threshold is searched to both sides, thereby accurately eliminating silent segments and transient pulse interference at the beginning and end of the signal. Pre-emphasis processing uses a transfer function of... A first-order high-pass filter enhances the energy of high-frequency components and balances spectral tilt. Framing operations divide the continuous signal into segments of 20 to 30 milliseconds, with a frame shift of 50% to ensure the continuity of feature extraction.

[0035] Step S3: Map the audio frame sequence to the Mel frequency domain and extract the voiceprint feature vector of each frame to construct the voiceprint feature matrix of the headphone to be detected.

[0036] like Figure 2 As shown, step S3 in this embodiment includes the following sub-steps: Step S301: Perform a Fast Fourier Transform (FFT) on each audio frame in the audio frame sequence to calculate the power spectral density function. For example, the number of points in the FFT is typically set to 512 or 1024, with zeros padded if necessary. The power spectral density reflects the energy distribution of the signal at various frequency points, and is calculated by dividing the square of the spectral amplitude by the number of transform points.

[0037] Step S302: The power spectral density function is filtered through a preset Mel filter bank to obtain the logarithmic Mel spectral energy value. In an optional embodiment, the Mel filter bank includes 26 to 40 triangular bandpass filters, whose center frequencies are uniformly distributed on the Mel scale, but exhibit a low-frequency density and high-frequency sparseness characteristic on the physical frequency scale. The logarithmic operation is used to simulate the nonlinear perception characteristics of loudness by the human ear, converting multiplicative noise into additive noise for easier subsequent processing.

[0038] Step S303: Perform a discrete cosine transform on the log-Mel spectrum energy value and retain the low-order coefficients as the voiceprint feature vector. The purpose of the discrete cosine transform is to remove the correlation between features of different dimensions and concentrate the energy on the low-frequency coefficients. Usually, the cepstral coefficients from the 2nd to the 13th dimension are retained as the voiceprint feature vector, while the 0th dimension and higher-order coefficients are discarded, because higher-order coefficients often contain small and rapid perturbations of the excitation source rather than the resonance characteristics of the vocal tract.

[0039] Step S4: Obtain the reference voiceprint model of the standard headphones. Based on the voiceprint feature matrix and the reference voiceprint model, calculate the voiceprint anomaly deviation index of the headphones to be tested using a weighted distance algorithm. For example, the reference voiceprint model is obtained by extracting voiceprint features from a large number of gold sample headphones that have been manually judged as qualified, and calculating their mean vector or training a Gaussian mixture model. It represents the standard acoustic fingerprint of the headphone model.

[0040] In step S4 of this embodiment, the formula for calculating the voiceprint anomaly deviation index is as follows: ,in, Indicates the voiceprint anomaly deviation index; This represents the total number of dimensions in the voiceprint feature vector; Indicates the headphone to be tested in the first... Feature coefficient values ​​in each feature dimension; Indicates standard headphones at the Reference feature coefficient values ​​in each feature dimension; Indicates the first The importance weighting coefficients for each feature dimension. A smaller voiceprint anomaly deviation index indicates that the sound quality is closer to the standard, while a larger value indicates more severe sound quality degradation. The calculation formula quantifies the degree of sound quality deviation of the tested earphone by summing the weighted squared differences of each feature dimension. This formula is essentially a weighted Euclidean distance in the feature space, which can also be replaced by Mahalanobis distance or cosine similarity distance.

[0041] The calculation formula quantifies the degree of sound quality deviation of the headphone under test by summing the weighted squared differences of each feature dimension.

[0042] Importance weight coefficient The setting logic is as follows: Obtain equal loudness curve data of human hearing; according to the first For each feature dimension, the corresponding actual physical frequency is used to find the corresponding auditory sensitivity value in the equal loudness curve data; the auditory sensitivity value is then normalized and assigned to the importance weight coefficient. This makes the characteristic differences in the frequency bands sensitive to human ears contribute more to the abnormal deviation index of voiceprints.

[0043] Step S5: Compare the voiceprint anomaly deviation index with a preset defect classification threshold, and determine the defect type and quality level of the earphone to be tested based on the comparison result. The preset threshold is determined through statistical analysis of historical defect sample data, for example, by taking the lower quartile of the historical defect deviation index distribution.

[0044] Step S5 includes: establishing a multi-level threshold table containing pure tone abnormality threshold, balance deviation threshold, and distortion threshold; if the voiceprint abnormality deviation index exceeds the pure tone abnormality threshold, the defect type is determined to be voice coil friction or diaphragm damage; if the voiceprint abnormality deviation index is less than the pure tone abnormality threshold but exceeds the balance deviation threshold, the defect type is determined to be left and right ear sensitivity mismatch.

[0045] The method in this embodiment also includes an enhanced detection step for transient abnormal sounds: in step S3, higher-order statistical features of the acoustic response signal are extracted, including at least the kurtosis coefficient and the skewness coefficient; the higher-order statistical features are fused with the voiceprint feature vector; and a support vector machine classifier is used to identify the fused features in order to detect minute abnormal sound defects of short-time pulse type.

[0046] like Figure 3 This embodiment also provides a headphone defect detection device based on voiceprint feature analysis. The device is typically integrated into an audio analyzer or an automated testing cabinet on a production line. The device includes: The acoustic data acquisition module is used to control the sound source to play a standard sweep frequency signal in an anechoic test environment and simultaneously acquire the acoustic response signal of the headphone under test. For example, the module includes a professional audio interface card that supports 48-volt phantom power supply, analog-to-digital conversion signal-to-noise ratio better than 100 dB, and total harmonic distortion plus noise less than -90 dB.

[0047] The signal optimization processing module is used to perform time-domain endpoint detection and frequency-domain noise reduction on the acoustic response signal to generate a clean audio frame sequence.

[0048] The speaker matrix construction module is used to map audio frame sequences to the Mel frequency domain and extract features to construct the speaker feature matrix of the headphone to be detected. This module performs the Mel frequency cepstral coefficient feature extraction process and supports hardware-accelerated FFT operations.

[0049] The anomaly index calculation module is used to call the pre-stored standard headphone reference voiceprint model to calculate the voiceprint anomaly deviation index of the headphone to be tested; this module performs weighted distance calculation and can use the SIMD instruction set for vectorization acceleration.

[0050] The defect grading and judgment module compares the voiceprint anomaly deviation index with a preset defect grading threshold to output the defect type and quality level of the headphones under test. The output results can drive the robotic arm of the production line to automatically sort good or bad products through a general input / output interface, or display the words "PASS" in green or "FAIL" in red on an LCD screen.

[0051] A computer-readable storage medium storing computer program instructions is characterized in that, when read and executed by one or more processors, the computer program instructions cause one or more processors to perform steps in a headphone defect detection method based on voiceprint feature analysis. In specific implementations, the storage medium stores not only binary program code for executing the detection logic but also a "standard headphone reference voiceprint model library," which contains pre-trained GMM model parameters or mean vector files for different headphone models and batches. Furthermore, the medium can be allocated a dedicated log partition for cyclically storing the most recent N detection result records, including test time, machine number, deviation index, and raw waveform data, to facilitate quality traceability and big data analysis. The storage medium includes, but is not limited to, industrial-grade CF cards, solid-state drives, embedded multimedia cards, or cloud storage endpoints connected to a factory MES system.

[0052] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for detecting headphone defects based on voiceprint feature analysis, characterized in that, The method includes the following steps: Step S1: Drive the earphone under test to play a standard sweep frequency signal in an anechoic test environment, and use an artificial ear to collect the acoustic response signal of the earphone under test; Step S2: Perform time-domain endpoint detection and frequency-domain noise reduction on the acoustic response signal to generate an audio frame sequence to be analyzed; Step S3: Map the audio frame sequence to the Mel frequency domain and extract the voiceprint feature vector of each frame to construct the voiceprint feature matrix of the headphone to be detected. Step S4: Obtain a reference voiceprint model of a standard headphone; and calculate the voiceprint anomaly deviation index of the headphone to be tested using a weighted distance algorithm based on the voiceprint feature matrix and the reference voiceprint model. Step S5: Compare the voiceprint abnormality deviation index with the preset defect classification threshold, and determine the defect type and quality level of the earphone to be tested based on the comparison result.

2. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, Step S3 includes the following sub-steps: Step S301: Perform a fast Fourier transform on each audio frame in the audio frame sequence to calculate the power spectral density function; Step S302: Filter the power spectral density function through a preset Mel filter bank to obtain the log-Mel spectrum energy value; Step S303: Perform a discrete cosine transform on the logarithmic Mel spectrum energy value and retain the low-order coefficients as the voiceprint feature vector.

3. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, In step S4, the formula for calculating the voiceprint anomaly deviation index is: ,in, This represents the voiceprint anomaly deviation index; This represents the total number of dimensions of the voiceprint feature vector; Indicates that the earphone to be tested is in the first... Feature coefficient values ​​in each feature dimension; Indicates that the standard headphones are in the first Reference feature coefficient values ​​in each feature dimension; Indicates the first The importance weight coefficients of each feature dimension; The calculation formula quantifies the degree of sound quality deviation of the headphone under test by accumulating the weighted squared differences of each feature dimension.

4. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, Step S1 includes configuring the frequency range of the standard sweep signal to cover 20 Hz to 2 Hz and setting the logarithmic sweep mode; during the acquisition process, environmental background noise data is recorded synchronously. An adaptive filter is used to remove the environmental background noise data from the acoustic response signal in order to eliminate the interference of the test environment on the voiceprint extraction.

5. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, Step S2 includes: calculating the short-time energy and short-time zero-crossing rate of the acoustic response signal; identifying the start and end points of the effective speech segments of the acoustic response signal based on the dual-threshold decision method of the short-time energy and the short-time zero-crossing rate; pre-emphasizing the extracted effective speech segments and performing frame-by-frame windowing operation using the Hamming window function to obtain the audio frame sequence.

6. The headphone defect detection method based on voiceprint feature analysis according to claim 3, characterized in that, The importance weight coefficient The setting logic is as follows: Obtain equal loudness curve data of human hearing; according to the first... The actual physical frequencies corresponding to each feature dimension are used to find the corresponding auditory sensitivity values ​​in the equal loudness curve data; these auditory sensitivity values ​​are then normalized and assigned to the importance weight coefficients. This makes the characteristic differences in the frequency bands sensitive to human ears contribute more to the voiceprint abnormality deviation index.

7. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, Step S5 includes: establishing a multi-level threshold table containing a pure tone abnormality threshold, a balance deviation threshold, and a distortion threshold; if the voiceprint abnormality deviation index exceeds the pure tone abnormality threshold, the defect type is determined to be voice coil friction or diaphragm damage; if the voiceprint abnormality deviation index is less than the pure tone abnormality threshold but exceeds the balance deviation threshold, the defect type is determined to be left and right ear sensitivity mismatch.

8. The headphone defect detection method based on voiceprint feature analysis according to claim 1, characterized in that, The method further includes an enhanced detection step for transient abnormal sounds: in step S3, higher-order statistical features of the acoustic response signal are extracted, the higher-order statistical features including at least kurtosis coefficient and skewness coefficient; the higher-order statistical features are fused with the voiceprint feature vector; and a support vector machine classifier is used to identify the fused features in order to detect minute abnormal sound defects of short-time pulse type.

9. A headphone defect detection device based on voiceprint feature analysis, characterized in that, The device includes: The acoustic data acquisition module is used to control the sound source to play a standard sweep frequency signal in an anechoic test environment and simultaneously acquire the acoustic response signal of the earphone under test. The signal optimization processing module is used to perform time-domain endpoint detection and frequency-domain noise reduction processing on the acoustic response signal to generate a clean audio frame sequence; The voiceprint matrix construction module is used to map the audio frame sequence to the Mel frequency domain and extract features to construct the voiceprint feature matrix of the headphone to be detected. The anomaly index calculation module is used to call the pre-stored standard headphone reference voiceprint model to calculate the voiceprint anomaly deviation index of the headphone to be tested. The defect classification and determination module is used to compare the voiceprint abnormality deviation index with the preset defect classification threshold to output the defect type and quality level of the earphone to be tested.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, The computer program instructions are read and executed by one or more processors, causing the one or more processors to perform the steps of the headphone defect detection method based on voiceprint feature analysis as described in any one of claims 1 to 8.

Images