Performance test method of audio system and electronic equipment

Abstract

The invention provides a performance testing method of an audio system and electronic equipment, and is applied to the technical field of audio testing. The performance test method of the audio system comprises the steps of obtaining audio information played by a target audio channel of the audio system in a to-be-tested product for a test information set, wherein the test information set comprises a preset tag of the audio information; extracting spectrum features in the audio information; classifying the spectrum features by using a preset neural network model to obtain classification tags corresponding to the spectrum features; and comparing the classification label with a preset label, and determining the performance state of the audio system according to a comparison result corresponding to the target audio channel.

Description

Technical Field

[0001] This disclosure relates to the field of audio testing technology, and in particular to a performance testing method and electronic device for an audio system. Background Technology

[0002] Audio system performance testing has become a crucial part of production line quality control. Existing testing methods mainly rely on manual listening or expensive shielded enclosure equipment, which suffers from problems such as high subjectivity, low efficiency, and poor environmental adaptability.

[0003] In the complex noise environment of a factory, existing automated solutions struggle to achieve high-precision and high-stability audio performance testing, failing to meet the requirements of large-scale production for testing efficiency and consistency. Summary of the Invention

[0004] In view of this, this disclosure provides a performance testing method and electronic device for an audio system.

[0005] According to the first aspect of this disclosure, a performance testing method for an audio system is provided. The method includes: acquiring audio information played by a target audio channel of the audio system in the product under test to a test information set, wherein the test information set contains preset labels for the audio information; extracting spectral features from the audio information; classifying the spectral features using a preset neural network model to obtain classification labels corresponding to the spectral features; comparing the classification labels with preset labels, and determining the performance status of the audio system based on the comparison results corresponding to the target audio channel.

[0006] According to embodiments of this disclosure, frequency cepstral processing is performed on audio information to obtain static features in the audio information; first-order difference processing and second-order difference processing are performed on the static features to obtain first-order dynamic features and second-order dynamic features; the first-order dynamic features, second-order dynamic features and static features are spliced ​​together to obtain spectral features.

[0007] According to embodiments of this disclosure, static features are concatenated with first-order dynamic features according to a first dimension to obtain intermediate features; intermediate features are concatenated with second-order dynamic features according to a second dimension to obtain spectral features, wherein the first dimension and the second dimension are different.

[0008] According to embodiments of this disclosure, static features are concatenated with first-order dynamic features according to a first dimension to obtain intermediate features; intermediate features are concatenated with second-order dynamic features according to a second dimension to obtain spectral features, wherein the first dimension and the second dimension are different.

[0009] According to embodiments of this disclosure, after extracting spectral features from audio information, the time domain interval of the spectral features is determined; if the time domain interval is greater than or equal to a set threshold, the spectral features are truncated or downsampled; if the time domain interval is less than the set threshold, the spectral features are filled.

[0010] According to embodiments of this disclosure, feature extraction of spectral features is performed through convolution operations to obtain composite features; the composite features are normalized to obtain normalized features; and the normalized features are subjected to linear transformation and nonlinear classification to obtain classification labels corresponding to the spectral features.

[0011] According to embodiments of this disclosure, the normalized features are subjected to dimensionality reduction processing to obtain dimensionality-reduced features, which include multiple sub-dimensionality-reduced features; the multiple sub-dimensionality-reduced features are subjected to classification processing to obtain the probability distribution of each sub-dimensionality-reduced feature on multiple preset classification labels; for each sub-dimensionality-reduced feature, the preset classification label corresponding to the maximum probability value in the probability distribution is used as the classification label.

[0012] According to embodiments of this disclosure, the spatial information where the audio information is played is determined; the category label is compared with a preset label associated with the spatial information; when the comparison result indicates that the category label is the same as the preset label, the comparison result is determined to be successful; when the comparison result indicates that the category label is different from the preset label, the comparison result is determined to be unsuccessful.

[0013] According to embodiments of this disclosure, the confidence level corresponding to the classification label output by the preset neural network model is obtained; when the comparison result indicates successful comparison and the confidence level is higher than the preset confidence level threshold, the target state of the target audio channel is determined to be normal; when the comparison result indicates failed comparison, or the comparison result indicates successful comparison but the confidence level is lower than the preset confidence level threshold, the target state of the target audio channel is determined to be abnormal; based on the target state of the target audio channel, the performance state of the audio system is determined.

[0014] According to embodiments of this disclosure, the target audio channel includes at least one of a left audio channel, a right audio channel, and a dual audio channel; when the target states of the left audio channel, the right audio channel, and the dual audio channel are all normal, the performance state of the audio system is determined to be normal; when the target state of any one of the left audio channel, the right audio channel, and the dual audio channel is abnormal, the performance state of the audio system is determined to be abnormal.

[0015] A second aspect of this disclosure provides an electronic device, comprising: a receiver configured to receive audio information played by a target audio channel of an audio system in a product under test to a specified test information set, the test information set containing preset labels for the audio information; and a processor communicatively connected to the receiver, the processor being configured to: extract spectral features from the audio information; classify the spectral features using a preset neural network model to obtain classification labels corresponding to the spectral features; compare the classification labels with preset labels, and determine the performance status of the audio system based on the comparison result corresponding to the target audio channel.

[0016] A third aspect of this disclosure provides a performance testing apparatus for an audio system, comprising: an acquisition module for acquiring audio information played by a target audio channel of the audio system in the product under test to a test information set, the test information set containing preset labels for the audio information; an extraction module for extracting spectral features from the audio information; a classification module for classifying the spectral features using a preset neural network model to obtain classification labels corresponding to the spectral features; and a comparison module for comparing the classification labels with preset labels and determining the performance status of the audio system based on the comparison results corresponding to the target audio channel.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0018] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 This illustration schematically depicts a performance testing method for an audio system and an application scenario of an electronic device according to embodiments of the present disclosure;

[0020] Figure 2 A flowchart illustrating a performance testing method for an audio system according to an embodiment of the present disclosure is shown schematically.

[0021] Figure 3 An overall flowchart illustrating the determination of spectral characteristics according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 4 A detailed flowchart illustrating the determination of spectral characteristics according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 5 This schematic diagram illustrates a structural block diagram of a performance testing apparatus for an audio system according to an embodiment of the present disclosure;

[0024] Figure 6A schematic block diagram of an image removal apparatus according to an embodiment of the present disclosure is shown. Detailed Implementation

[0025] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely examples and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "having at least one of A, B and C" should include, but is not limited to, having only A, having only B, having only C, having A and B, having A and C, having B and C, and / or having A, B and C, etc.).

[0029] Before introducing the technical solutions provided in the embodiments of this disclosure, the relevant technologies involved in this disclosure will be explained first.

[0030] In related technologies, the performance testing of audio systems mainly faces two key challenges: First, traditional manual testing methods rely on the subjective judgment of operators, which has problems such as low efficiency, poor consistency, and susceptibility to fatigue. Second, existing automated solutions based on general speech recognition models perform poorly in complex noise environments in factories, lacking sufficient sensitivity to detect sound quality defects such as speaker distortion and hoarseness, and failing to effectively distinguish between environmental noise and functional abnormalities of the equipment itself.

[0031] Specifically, in the actual production line environment, existing testing solutions often rely solely on the accuracy of speech-to-text conversion to determine speaker status. This method is insensitive to subtle changes in audio quality and struggles to capture abnormal patterns at the spectral feature level. Furthermore, general-purpose models lack optimization for specific test vocabulary sets, making them prone to misjudgments in complex acoustic environments such as multi-voice and mechanical noise environments. Consequently, the reliability and stability of test results fail to meet the demands of large-scale production. It is particularly noteworthy that existing solutions typically use continuous speech with a large vocabulary as test material. This type of test content itself introduces additional recognition complexity, hindering the accurate detection of core performance indicators of the audio system.

[0032] In view of this, embodiments of the present disclosure provide a performance testing method for an audio system, applicable to the field of audio testing. The method includes: acquiring audio information played by a target audio channel of the audio system in the product under test to a test information set; the test dataset uses a carefully selected small-scale vocabulary set, with each audio information corresponding to a precise preset label; extracting spectral features from the audio information; classifying the spectral features using a preset neural network model to obtain classification labels corresponding to the spectral features; comparing the classification labels with preset labels, and determining the performance status of the audio system based on the comparison results corresponding to the target audio channel.

[0033] This embodiment of the disclosure, by employing a limited test vocabulary set and combining Mel-frequency cepstral coefficient feature extraction with convolutional neural network classification and recognition specifically optimized for production line environments, effectively overcomes the subjectivity of manual testing and the adaptability of general models in noisy environments. The precise test dataset used in this method significantly reduces environmental noise interference, allowing the model to focus on the feature analysis of the audio quality itself. It can not only accurately identify the playback content but also effectively detect sound quality anomalies through spectral feature analysis. This achieves automated, high-precision testing of audio system performance in complex, noisy open-space environments, providing a reliable unmanned testing solution for production lines.

[0034] This disclosed embodiment can be widely applied to the speaker production line testing of electronic devices such as mobile phones, tablets, laptops, smart speakers, smartwatches, and automotive audio systems, and is particularly suitable for automated quality inspection in large-scale factory production environments. Furthermore, this method can also be extended to scenarios such as functional diagnostics in consumer electronics repair centers and batch acceptance processes for multimedia equipment, providing a unified and reliable evaluation standard for audio system performance testing.

[0035] Figure 1 The schematic diagram illustrates a flow chart of a performance testing method for an audio system according to an embodiment of the present disclosure.

[0036] like Figure 1As shown, the application scenario 100 according to this embodiment may include a test device 101, a production line 102 for the device under test, a test server 103, and a network 104 connecting the various entities.

[0037] Test device 101 is a dedicated device deployed at the test station on the production line, used to collect audio information played by the device under test (DUT) on production line 102 within an open space. Test device 101 is equipped with a high-precision audio acquisition interface and computing unit, possessing independent signal processing and neural network inference capabilities, and can complete the entire process of audio acquisition, feature extraction, and classification judgment on-site.

[0038] The equipment under test (UDT) production line 102 includes multiple UDTs, which are located in the same open acoustic environment as the test equipment 101. Signal transmission between the two is achieved through acoustic field coupling, and there is no physical electrical connection between them. During testing, the audio system of the UDT on the production line 102 plays audio signals according to a preset test dataset.

[0039] Test server 103 serves as the model management and data service center, establishing a communication connection with test device 101 via network 104. Test server 103 is responsible for the training optimization, version management, and centralized distribution of preset neural network models.

[0040] Network 104 includes, but is not limited to, industrial Ethernet, wireless LAN, or wired network, and is used to establish a data channel between test equipment 101 and test server 103. This network connection is mainly used for the following scenarios: 1) periodically updating the preset neural network model parameters in test equipment 101; 2) synchronously updating the test dataset and preset label information; 3) collecting statistical data and test results from each test station.

[0041] It should be noted that although the test device 101 and the test server 103 are connected through the network 104, when performing a single test task, the test device 101 can independently complete all real-time processing steps, including audio acquisition, feature extraction, model inference and result judgment, without relying on the real-time performance of the network connection.

[0042] Accordingly, the audio system performance testing method provided in this embodiment is mainly implemented through the testing equipment 101, including: acquiring audio information played by the device under test on the production line 102 of the device under test through the built-in audio acquisition device; extracting the spectral features of the audio information; classifying the features using a locally deployed neural network model to obtain classification labels; and comparing the classification labels with preset labels to determine the system performance status.

[0043] It should be understood that Figure 1The system architecture shown can be adjusted according to actual needs. The test device 101 can be deployed as a distributed independent node or configured as a terminal device that relies on a server for real-time computing. The performance testing method of the audio system according to the embodiments of this disclosure will be described in detail below.

[0044] Figure 2 A flowchart illustrating a performance testing method for an audio system according to an embodiment of the present disclosure is shown.

[0045] like Figure 2 As shown, the performance testing method for the audio system in this embodiment can be applied to a production line testing system. The performance testing method for the audio system includes operations S210 to S240.

[0046] In operation S210, the audio information played by the target audio channel of the audio system in the product under test to the test information set is obtained.

[0047] In this embodiment of the disclosure, the product under test refers to an electronic device that requires audio performance testing in a production line environment. Taking a smart speaker as an example, the audio system of a smart speaker needs to ensure that basic voice playback functions are normal. In a large-scale production environment with a monthly output of tens of thousands of units, the product under test sequentially passes through testing stations to complete automated testing.

[0048] The audio system refers to the complete hardware chain in the product under test (DUT) that enables audio playback, including audio codec chips, power amplifier circuits, and speaker units. The audio system in the DUT may experience functional problems such as distortion or hoarseness due to poor assembly or malfunctions, such as loose connectors or component failures.

[0049] The target audio channel refers to the audio path that can be evaluated independently during testing. The target audio channel may include the left channel, right channel, and a stereo mix. The test information set refers to a limited vocabulary audio dataset specifically designed to suit the production line environment.

[0050] For example, an audio dataset can contain around 20 words. The number of words in the audio dataset can be configured according to the complexity of the testing environment, ranging from 18 to 30 words. Words in the audio dataset are unique, and each word is recorded more than 1000 times using standard equipment in a silent laboratory environment. The audio information refers to sound wave signals collected via microphone in an open production line environment. The sound wave signals include the test audio played by the product under test, background noise, such as the operating sounds of adjacent equipment and personnel conversations. The data collection process does not rely on soundproof enclosures and is completed in a real production line environment.

[0051] In operation S220, spectral features are extracted from the audio information.

[0052] In this embodiment of the disclosure, the extracted spectral features are multi-dimensional feature vectors constructed based on the Mel frequency cepstral system. By fusing the static spectral characteristics and dynamic evolution laws of the audio signal, the feature vectors can characterize abnormal sound quality patterns and provide neural network classification with feature inputs that are sensitive to faults such as crackling and hoarseness.

[0053] In operation S230, a preset neural network model is used to classify the spectral features and obtain the classification labels corresponding to the spectral features.

[0054] In this embodiment, the preset neural network model is a deep learning network specifically trained on audio data from a production line environment, possessing the ability to recognize audio features in complex acoustic environments. Classification refers to the computational process by which the neural network model performs pattern recognition and analysis on the input spectral features. By analyzing the similarity between the feature vectors in the spectral features and various audio templates, it outputs the probability distribution across all preset categories. The classification label can be seen as the result determined by the preset neural network model, representing the identified test word category.

[0055] In operation S240, the classification labels are compared with preset labels, and the performance status of the audio system is determined based on the comparison results corresponding to the target audio channel.

[0056] In this embodiment, the preset label refers to a standard identifier pre-defined for each audio sample in the test dataset, representing the target words expected to be correctly identified during audio playback. The comparison result is a judgment conclusion obtained by matching the classification label output by the neural network with the preset label, and the judgment conclusion can be divided into comparison success and comparison failure. The performance status is an evaluation conclusion of the audio system's working condition based on the comparison result, and the performance status reflects the audio system's ability to achieve the expected playback effect under the target audio channel.

[0057] In this embodiment, audio information played by the audio system of the product under test on a test dataset is obtained. After extracting the spectral features, a pre-defined neural network model is used for classification to obtain classification labels. The classification labels are compared with pre-defined labels, and the performance status of the audio system is determined based on the comparison results. A complete automated testing process is constructed. Through spectral feature extraction and neural network classification, objective evaluation of the audio system performance is achieved, replacing traditional subjective human judgment. This significantly improves testing efficiency and consistency, providing a feasible technical foundation for large-scale application on production lines and enabling automated testing of loudspeakers in complex, noisy open-space environments.

[0058] The following provides a detailed description of the performance testing methods for the audio system, including operations S210 to S240.

[0059] In this embodiment of the disclosure, the audio information played by the target audio channel of the audio system in the product under test for the test information set in operation S210 can be obtained in the following ways: On a real production line, a noise-canceling microphone array is deployed at the test station to simultaneously collect the audio signal played by the product under test and the ambient background noise in an open environment. Alternatively, for a multi-channel audio system, a multi-input channel acquisition device is used to simultaneously acquire the audio signals of each target audio channel, such as the left and right channels.

[0060] The process of extracting spectral features from audio information will now be explained.

[0061] Figure 3 An overall flowchart illustrating the determination of spectral characteristics according to an embodiment of the present disclosure is shown.

[0062] like Figure 3 As shown, the above operation S220 may also include operations S301 to S303.

[0063] In operation S301, frequency cepstral processing is performed on the audio information to obtain the static features in the audio information.

[0064] In this embodiment of the disclosure, the audio information is processed by frame segmentation, dividing the audio information into continuous short audio frames; for each audio frame, pre-emphasis, windowing, and Fourier transform are performed sequentially to obtain the spectrum; the spectrum is filtered by a Mel filter bank and logarithmic operation is performed; the obtained logarithmic Mel spectrum is subjected to discrete cosine transform to extract the static cepstral coefficients that characterize the spectral envelope of the audio signal, which are used as static features of the audio information.

[0065] For example, the audio information is constructed from about 20 carefully selected words to form a test dataset. Each word is recorded more than 1,000 times to ensure data diversity. The test dataset can be waveform files of word audio. After reading the audio waveform file, the audio waveform file is extracted and processed by Mel-frequency cepstral coefficients (MFCC) and then used as static features.

[0066] In operation S302, the static features are subjected to first-order and second-order difference processing respectively to obtain first-order dynamic features and second-order dynamic features.

[0067] In this embodiment, first-order differential processing refers to obtaining dynamic features characterizing short-term spectral change trends by calculating the difference between static features of adjacent frames; second-order differential processing further calculates the difference based on first-order differential processing to obtain dynamic features characterizing spectral change acceleration. The dynamic features characterizing short-term spectral change trends and the dynamic features characterizing spectral change acceleration can effectively capture the transient characteristics of audio signals.

[0068] In operation S303, the first-order dynamic features, second-order dynamic features, and static features are spliced ​​together to obtain the spectral features.

[0069] In this embodiment, the static features representing the spectral envelope, the first-order dynamic features representing short-term changes, and the second-order dynamic features representing the acceleration of change are connected and combined along the dimension of their feature coefficients to determine a more comprehensive fusion feature that combines static structure and dynamic temporal information. This fusion feature is then used as the spectral feature, where the fusion feature is the input data of a preset neural network model.

[0070] For example, static features can be Mel-frequency cepstral coefficients, first-order dynamic features can be the first-order difference coefficients of the Mel-frequency cepstral coefficients, and second-order dynamic features can be the second-order difference coefficients of the Mel-frequency cepstral coefficients. The specific operation of concatenation involves linking the Mel-frequency cepstral coefficients, first-order difference coefficients, and second-order difference coefficients corresponding to the same time frame in a sequential manner along the feature dimension, forming a fused feature vector. This fused feature vector simultaneously contains the static spectral characteristics, short-term dynamic change trends, and acceleration information of the audio signal, significantly improving the representational ability of the features and providing richer discriminative basis for the accurate classification of subsequent neural network models.

[0071] In this embodiment, static features are extracted through frequency cepstral processing, and dynamic features are obtained by combining first-order and second-order difference processing. The static features and the two dynamic features are then concatenated to form spectral features. This effectively integrates the static spectral characteristics of audio with its first-order and second-order dynamic change information, effectively improving the completeness and discriminative power of feature representation, and providing richer feature inputs for subsequent neural network classification.

[0072] Figure 4 A detailed flowchart illustrating the determination of spectral characteristics according to an embodiment of the present disclosure is shown schematically.

[0073] like Figure 4 As shown, the above operation S303 may also include operations S401 to S402.

[0074] In operation S401, the static features and the first-order dynamic features are concatenated according to the first dimension to obtain the intermediate features.

[0075] In this embodiment of the disclosure, the first dimension is the feature channel dimension. By connecting all the coefficients of the static feature and all the coefficients of the first-order dynamic feature along the feature channel direction, an intermediate feature is formed that simultaneously contains the original static characteristics and the first-order dynamic change information.

[0076] In operation S402, the intermediate features and the second-order dynamic features are concatenated according to the second dimension to obtain the spectral features.

[0077] In this embodiment, the second dimension is the time frame dimension. By connecting and combining intermediate features and second-order dynamic features along the time frame dimension, a spectral feature is constructed that simultaneously includes static features, first-order dynamic features, and second-order dynamic features. The first and second dimensions are orthogonal to each other, achieving feature fusion from two different perspectives: feature channels and temporal evolution. In this embodiment, a step-by-step splicing strategy is adopted. First, static features and first-order dynamic features are spliced ​​along the first dimension to obtain intermediate features. Then, intermediate features and second-order dynamic features are spliced ​​along different second dimensions to obtain spectral features. By splicing features along different dimensions, feature confusion is avoided, the integrity of multi-scale features is ensured, and the efficiency and effectiveness of feature fusion are improved.

[0078] In this embodiment of the disclosure, after the operation S402 extracts the spectral features from the audio information, it may further include: determining the time domain interval of the spectral features; if the time domain interval is greater than or equal to a set threshold, then truncating or downsampling the spectral features; if the time domain interval is less than the set threshold, then filling the spectral features.

[0079] In this embodiment, the truncation operation specifically involves removing tail feature frames exceeding a set threshold along the time dimension, while retaining the spectral features of a consecutive set threshold number of frames starting from the initial time. The downsampling operation employs a mean pooling method, compressing the length of the spectral features to within the set threshold range by locally averaging the features of multiple adjacent frames along the time dimension at a fixed ratio. The padding operation uses a cyclic copying method, using the last frame of the spectral features as a padding template and repeatedly adding it to the tail of the spectral features as needed, until the total number of frames of the spectral features reaches the set threshold requirement.

[0080] For example, when the threshold is set to 300 frames, for input features of 350 frames, the truncation operation directly discards the last 50 frames; the downsampling operation can perform mean pooling on every 7 frames to generate 1 frame, realizing the conversion from 350 frames to 300 frames. For input features of 250 frames, the last 50 frames are selected as the filling template and copied once in a loop to make the total number of frames reach 300. All processing keeps the feature dimension unchanged, only adjusting the time axis length.

[0081] It should be noted that the truncation, downsampling, and padding operations all maintain the frequency dimension of the spectral features unchanged, adjusting only the length of the time axis. The spectral features, after truncation and padding, are then transposed to adjust the feature dimensions to match the input requirements of the neural network model for its use.

[0082] In this embodiment, after extracting spectral features, the relationship between the time domain interval and a set threshold is determined. Features that are too long are truncated or downsampled, while features that are too short are padded. This standardization of spectral features ensures that audio inputs of different durations can meet the fixed-dimensional requirements of the neural network model, improving system robustness.

[0083] In this embodiment of the disclosure, operation S230 may further include: extracting features from the spectral features through convolution operations to obtain composite features; normalizing the composite features to obtain normalized features; and performing linear transformation and nonlinear classification processing on the normalized features to obtain classification labels corresponding to the spectral features.

[0084] In this embodiment, deep feature extraction of the input spectral features is performed through multi-layer two-dimensional convolution operations. The convolution kernel slides along the feature dimension and time dimension, and composite features containing information at different levels are gradually abstracted through multiple convolutional layers. Subsequently, batch normalization processing is performed on the composite features. Data distribution is standardized by calculating the mean and variance of the features within the batch, and normalized features with consistent scale are obtained. Finally, the normalized features are input into a fully connected layer for linear transformation, and non-linear classification processing is performed through a normalization exponential function to output the probability distribution corresponding to each preset word category. The category corresponding to the highest probability value is taken as the classification label corresponding to the spectral feature.

[0085] For example, a convolutional neural network with six 2D convolutional layers is used to perform deep feature extraction on spectral features to obtain composite features. The composite features are then batch normalized by a processing module with six normalization layers, and discard regularization is used to randomly drop neuron connections. Linear transformation is performed through two fully connected layers, and nonlinear classification is completed by combining a normalized exponential function. This maps the features to the corresponding 20 word classification labels. This process ensures the model training speed and stability while effectively preventing overfitting and significantly improving classification accuracy and generalization ability.

[0086] In this embodiment of the disclosure, the normalized features are subjected to dimensionality reduction processing to obtain dimensionality-reduced features, wherein the dimensionality-reduced features include multiple sub-dimensionality-reduced features; the multiple sub-dimensionality-reduced features are subjected to classification processing to obtain the probability distribution of each sub-dimensionality-reduced feature on multiple preset classification labels; for each sub-dimensionality-reduced feature, the preset classification label corresponding to the maximum probability value in the probability distribution is used as the classification label.

[0087] In this embodiment, the dimensionality reduction process projects normalized features into a low-dimensional space through a fully connected layer, forming multiple independent feature sub-vectors as sub-dimensionality reduction features. Each sub-dimensionality reduction feature is processed by an independent classifier, which calculates the probability distribution of the sub-feature across all preset classification labels using a fully connected layer and a normalized exponential function. By independently judging the probability distributions output by each sub-dimensionality reduction feature, the preset classification label corresponding to the maximum probability value in each distribution is selected as the classification result of that sub-feature. Finally, the system integrates the classification results of all sub-dimensionality reduction features and determines the final classification label output using a voting mechanism or a weighted fusion method. This multi-feature parallel classification mechanism effectively improves the model's classification robustness under different acoustic scenarios.

[0088] In this embodiment, deep composite features of spectral features are extracted through convolution operations. After normalization, the normalized features are then subjected to dimensionality reduction to obtain multiple sub-dimensionality-reduced features. Subsequently, each sub-dimensionality-reduced feature undergoes linear transformation and nonlinear classification processing to obtain its probability distribution on a preset classification label. The classification label corresponding to the highest probability value is selected as the classification result. A complete neural network classification process is constructed. The feature extraction capability is fully utilized through convolution operations, and the stability of features is improved by combining normalization processing. Furthermore, multiple sub-dimensionality-reduced features are obtained through dimensionality reduction processing, and classification decisions are made based on probability distribution analysis, effectively improving classification accuracy and result reliability.

[0089] In this embodiment of the disclosure, the operation S240 of comparing the classification label with the preset label may further include: determining the spatial information where the audio information is played; comparing the classification label with the preset label associated with the spatial information; when the comparison result indicates that the classification label is the same as the preset label, the comparison result is determined to be successful; when the comparison result indicates that the classification label is different from the preset label, the comparison result is determined to be unsuccessful.

[0090] In this embodiment of the disclosure, spatial information includes spatial orientation information and spatial noise information. Spatial orientation information refers to the physical orientation of the sound source relative to the acquisition device during audio playback, including specific orientations such as left channel, right channel, front, and rear. For example, when testing the left channel, the expected audio signal should mainly come from the left side. Spatial noise information refers to the background noise characteristics present in the acquisition environment, including noise type (such as production line equipment operation noise, human voice), noise intensity (decibels), and noise spectrum characteristics.

[0091] Preset labels associated with spatial information refer to the label mapping relationships established during the model training phase for different spatial scenarios. For spatial orientation, the same test word corresponds to different preset labels under different playback orientations. For noisy environments, the same test word corresponds to different preset labels under different noise scenarios.

[0092] In this embodiment of the disclosure, the spatial orientation and noise characteristics of the current test environment are identified, a preset set of labels matching the spatial information is selected, and the classification labels output by the neural network are matched with the selected preset labels. For example, in the left channel test mode, the recognition result is compared with the words in the left channel label set, but not with the expected words in the right channel.

[0093] It should be noted that the spatial scene comparison mechanism has significant value in production line environments. For example, when testing multiple devices simultaneously in an open, noisy environment, spatial orientation information can be used to accurately denoise spatial noise and identify units that do not correspond to the target device. This mechanism significantly improves the identification accuracy, system anti-interference capability, and overall testing efficiency in multi-device parallel testing scenarios.

[0094] In this embodiment of the disclosure, the combination of spatial orientation and noise characteristics of the current test environment is identified, a preset label set matching the composite spatial information is selected, and the classification labels output by the neural network are matched with the selected preset labels. For example, in the test mode of right channel and production line equipment operating noise, the identification result is compared with a label set specifically trained for this composite scenario.

[0095] It should be noted that the comparison mechanism based on multi-dimensional spatial scenarios fully considers the complexity of real test environments, further improving the accuracy and environmental adaptability of the tests.

[0096] In this embodiment, preset labels are associated with spatial information. By determining the spatial information of audio playback, the classification labels are compared with the preset labels corresponding to the spatial information to determine the comparison result. This scheme introduces spatial dimension verification, which can effectively detect spatially related audio system faults such as reversed channel connections and incorrect orientation.

[0097] In this embodiment of the disclosure, the process of determining the performance status of the audio system based on the comparison result corresponding to the target audio channel in operation S240 may further include: obtaining the confidence level corresponding to the classification label output by the preset neural network model; when the comparison result indicates successful comparison and the confidence level is higher than a preset confidence level threshold, determining the target status of the target audio channel as normal; when the comparison result indicates failed comparison, or the comparison result indicates successful comparison but the confidence level is lower than a preset confidence level threshold, determining the target status of the target audio channel as abnormal; and determining the performance status of the audio system based on the target status of the target audio channel.

[0098] In this embodiment, a confidence level mechanism is introduced to evaluate the performance status of the audio system. Specifically, the confidence level refers to the degree of certainty of the classification result by the neural network model, quantified by the maximum probability value output by the normalized exponential function; the confidence level threshold is a judgment boundary value determined based on statistical analysis of historical test data. When the classification label matches the preset label and the confidence level exceeds the threshold, it indicates that the target audio channel is working normally; when there is a label mismatch or insufficient confidence level, it is judged as an abnormal state. For multi-channel systems, a veto mechanism is adopted, and any abnormality in any channel is judged as an overall system abnormality.

[0099] It should be noted that the classification labels output by the preset neural network model can be either word names or folder names in the dataset. That is, different word names and Chinese folder names have their corresponding confidence levels.

[0100] For example, confidence level refers to the maximum probability value generated by the normalized exponential function in the output layer of the neural network, with a value ranging from 0 to 1; the confidence level threshold is set to 0.95 based on the statistical results of production line test data. When the classification label matches the preset label and the corresponding confidence level is greater than or equal to 0.95, the target audio channel is considered to be working normally; when the classification label does not match the preset label, or although the label matches but the confidence level is lower than 0.95, the target audio channel is considered to be abnormal. For multi-channel audio systems, when any of the left, right, or dual audio channels is determined to be abnormal, the entire audio system's performance status is determined to be abnormal.

[0101] It should be noted that the confidence threshold (e.g., 0.95) is set based on statistical analysis of a large amount of actual test data. Validation data shows that a properly functioning speaker maintains a consistently high confidence level for word recognition (e.g., 1.0), while the confidence level drops significantly below this threshold when abnormalities such as distortion or hoarseness occur. Therefore, using this threshold for judgment can effectively distinguish system states and ensure the accuracy of performance status assessment.

[0102] In this embodiment of the disclosure, the test data comparison results of manual testing and neural network automatic testing of the audio system in the product under test on the production line are shown in Table 1:

[0103] Table 1

[0104]

[0105] The information recorded in Table 1 is explained below. Before the adoption of the neural network automatic testing method, from January to February 202x, the total number of tests was about 185,000 units, using purely manual testing. Although the pass rate was 0, there were problems such as limited testing scale and high labor costs. After the adoption of the neural network automatic testing method based on confidence threshold from March to June 202x, the total number of tests increased significantly to more than 468,000 units, and no quality complaints occurred.

[0106] In terms of efficiency improvement, the number of tests requiring manual review has been significantly reduced from approximately 100,000 units per month before implementation to approximately 2,000 units per month, achieving significant manpower savings and demonstrating the high efficiency of the confidence-based judgment mechanism in claim 8 on the actual production line. Quality data shows that the over-kill rate of AI testing has remained stable at around 1.5%, consistent with early test data, indicating that the neural network classification model has stable judgment characteristics; while its true defect rate has increased from 0.0168% before implementation to 0.0201%, not only not decreasing but also improving, confirming that this disclosure can more effectively detect potential defective products.

[0107] In this embodiment of the disclosure, determining the performance state of the audio system based on the target state of the target audio channel may further include: when the target states of the left audio channel, the right audio channel, and the dual audio channels are all normal, the performance state of the audio system is determined to be normal; when the target state of any one of the left audio channel, the right audio channel, and the dual audio channels is abnormal, the performance state of the audio system is determined to be abnormal.

[0108] In this embodiment, the target audio channel includes at least one of a left audio channel, a right audio channel, and a dual audio channel. The left audio channel corresponds to the independent testing of the left speaker unit of the product under test, the right audio channel corresponds to the independent testing of the right speaker unit, and the dual audio channel corresponds to a mixed test mode in which the left and right channels work simultaneously. Performance status determination adopts a full-channel consistency principle: the overall performance status of the audio system is ultimately determined to be normal only when the target status of all three audio channels (left, right, and dual) is determined to be normal; if any one of the three channels (left, right, and dual) is determined to be abnormal, the overall performance status of the audio system is immediately determined to be abnormal.

[0109] For example, during production line testing, the system sequentially tests the left, right, and dual audio channels: when the test word is correctly recognized in the left channel with a confidence level of 1.0, the test word is correctly recognized in the right channel with a confidence level of 1.0, and the test word is correctly recognized in both channels with a confidence level of 1.0, the system determines that all channels are functioning normally, and ultimately determines that the audio system's performance is normal. Conversely, if the word recognition result of any channel is inconsistent with expectations during the test (e.g., recognizing "seven" as "eleven"), or if the recognition confidence level drops significantly (e.g., below 0.95), the system immediately determines that the channel is abnormal, and thus determines that the overall audio system performance is abnormal. This judgment mechanism based on multi-channel independent testing can effectively detect typical problems such as single-sided speaker failure and channel imbalance.

[0110] In this embodiment of the disclosure, the audio information played by the target audio channel of the audio system in the product under test to the test information set can be subjected to multiple neural network tests to determine the performance status of the audio system in the product under test. The results of multiple tests are shown in Table 2:

[0111] Table 2

[0112]

[0113] It should be noted that in the large-scale production line test of the factory with a total of 108,923 units, by extracting the spectral features of the audio system of the product under test and classifying them using a preset neural network model, and comparing the classification labels with preset labels to determine the performance status, a final pass rate of 98.47% was achieved. Among them, the first test pass rate reached 95.60%, proving that the spectral feature extraction and classification method is efficient. The second pass rate of 2.49% and the third pass rate of 0.38% reflect the effectiveness of the confidence judgment and multi-round verification mechanism. Moreover, the results of manual review and automatic judgment for each model are consistent, verifying the reliability and stability of the method on different product models.

[0114] Figure 5 A schematic block diagram of a performance testing apparatus for an audio system according to an embodiment of the present disclosure is shown.

[0115] like Figure 5 As shown, the audio system performance testing device 500 of this embodiment includes an acquisition module 510, an extraction module 520, a classification module 530, and a comparison module 540.

[0116] The acquisition module 510 acquires the audio information played by the target audio channel of the audio system in the product under test to the test information set, which contains preset labels for the audio information. The extraction module 520 extracts the spectral features from the audio information. The classification module 530 classifies the spectral features using a preset neural network model to obtain the classification labels corresponding to the spectral features. The comparison module 540 compares the classification labels with preset labels and determines the performance status of the audio system based on the comparison results corresponding to the target audio channel.

[0117] According to embodiments of this disclosure, any plurality of modules among the acquisition module 510, encoding module 520, output module 530, and decoding module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the acquisition module 510, encoding module 520, output module 530, and decoding module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 510, encoding module 520, output module 530, and decoding module 540 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0118] Figure 6 A block diagram schematically illustrating an electronic device suitable for implementing performance testing according to an embodiment of the present disclosure is shown.

[0119] like Figure 6 As shown, an electronic device 600 according to an embodiment of this disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this disclosure.

[0120] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 602 and / or RAM 603. It should be noted that programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of the present disclosure by executing programs stored in one or more memories.

[0121] According to embodiments of this disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 608 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0122] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0123] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0124] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.

[0125] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0126] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0127] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0128] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0130] Those skilled in the art will understand that the features described in the various embodiments of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0131] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A performance testing method for an audio system, the method comprising: Obtain the audio information played by the target audio channel of the audio system in the product under test to the test information set, wherein the test information set contains preset tags for the audio information; Extract the spectral features from the audio information; The spectral features are classified using a preset neural network model to obtain the classification labels corresponding to the spectral features; The classification labels are compared with the preset labels, and the performance status of the audio system is determined based on the comparison results corresponding to the target audio channel.

2. The method according to claim 1, wherein extracting the spectral features from the audio information comprises: The audio information is subjected to frequency cepstral processing to obtain the static features in the audio information; The static features are subjected to first-order and second-order difference processing respectively to obtain first-order dynamic features and second-order dynamic features. The first-order dynamic feature, the second-order dynamic feature, and the static feature are concatenated to obtain the spectral feature.

3. The method according to claim 2, wherein concatenating the first-order dynamic feature, the second-order dynamic feature, and the static feature to obtain the spectral feature comprises: The static feature and the first-order dynamic feature are concatenated according to the first dimension to obtain the intermediate feature; The intermediate features are concatenated with the second-order dynamic features according to the second dimension to obtain the spectral features, wherein the first dimension and the second dimension are different.

4. The method according to claim 1, further comprising, after extracting the spectral features from the audio information: Determine the time domain interval of the spectral characteristics; If the time domain interval is greater than or equal to a set threshold, the spectral features are truncated or downsampled. If the time domain interval is smaller than the set threshold, then the spectral features are filled.

5. The method according to claim 1, wherein classifying the spectral features using a pre-trained neural network model to obtain the classification labels corresponding to the spectral features includes: The spectral features are extracted using convolution operations to obtain composite features; The composite features are normalized to obtain normalized features; The normalized features are subjected to linear transformation and nonlinear classification processing to obtain the classification labels corresponding to the spectral features.

6. The method according to claim 5, wherein performing linear transformation and nonlinear classification processing on the normalized features to obtain the classification label corresponding to the spectral features includes: The normalized features are subjected to dimensionality reduction processing to obtain dimensionality-reduced features, wherein the dimensionality-reduced features include multiple sub-dimensionality-reduced features; The multiple sub-dimensionality reduction features are classified to obtain the probability distribution of each sub-dimensionality reduction feature on multiple preset classification labels; For each of the sub-dimensionality reduction features, the preset classification label corresponding to the maximum probability value in the probability distribution is used as the classification label.

7. The method according to claim 1, wherein the preset tag is associated with the spatial information of the audio information being played; The step of comparing the category label with the preset label includes: Determine the spatial information in which the audio information is played; The classification label is compared with a preset label associated with the spatial information. When the comparison result indicates that the classification label is the same as the preset label, the comparison result is determined to be a successful comparison. When the comparison result indicates that the classification label is different from the preset label, the comparison result is determined to be a comparison failure.

8. The method according to claim 1, wherein comparing the classification label with the preset label and determining the performance status of the audio system based on the comparison result corresponding to the target audio channel includes: Obtain the confidence level corresponding to the classification label output by the preset neural network model; When the comparison result indicates a successful comparison and the confidence level is higher than the preset confidence level threshold, the target state of the target audio channel is determined to be a normal state. When the comparison result indicates a failure, or when the comparison result indicates a successful comparison but the confidence level is lower than a preset confidence level threshold, the target state of the target audio channel is determined to be an abnormal state. The performance status of the audio system is determined based on the target status of the target audio channel.

9. The method according to claim 8, wherein the target audio channel includes at least one of a left audio channel, a right audio channel, and dual audio channels; Determining the performance status of the audio system based on the target status of the target audio channel includes: When the target states of the left audio channel, the right audio channel, and the dual audio channels are all normal, the performance state of the audio system is determined to be normal. When the target state of any one of the left audio channel, the right audio channel, and the dual audio channels is abnormal, the performance state of the audio system is determined to be abnormal.

10. An electronic device, comprising: The receiver is configured to receive audio information played by a target audio channel of the audio system in the product under test to a specified test information set, wherein the test information set contains preset tags for the audio information. A processor, communicatively connected to the receiver, is configured to: Extract the spectral features from the audio information; The spectral features are classified using a preset neural network model to obtain the classification labels corresponding to the spectral features; The classification labels are compared with the preset labels, and the performance status of the audio system is determined based on the comparison results corresponding to the target audio channel.

Images