WO2021248522A1 - 气流杂音检测方法、装置、终端及存储介质 - Google Patents

气流杂音检测方法、装置、终端及存储介质 Download PDF

Info

Publication number
WO2021248522A1
WO2021248522A1 PCT/CN2020/096685 CN2020096685W WO2021248522A1 WO 2021248522 A1 WO2021248522 A1 WO 2021248522A1 CN 2020096685 W CN2020096685 W CN 2020096685W WO 2021248522 A1 WO2021248522 A1 WO 2021248522A1
Authority
WO
WIPO (PCT)
Prior art keywords
audio
audio signal
airflow noise
machine learning
sample
Prior art date
Application number
PCT/CN2020/096685
Other languages
English (en)
French (fr)
Inventor
吴锐兴
田晓晖
叶利剑
Original Assignee
瑞声声学科技(深圳)有限公司
瑞声科技(新加坡)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 瑞声声学科技(深圳)有限公司, 瑞声科技(新加坡)有限公司 filed Critical 瑞声声学科技(深圳)有限公司
Publication of WO2021248522A1 publication Critical patent/WO2021248522A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Definitions

  • This application relates to the field of computer technology, and in particular to a method, device, terminal, and storage medium for detecting airflow noise.
  • the cavity of the micro speaker is narrow, and the air flow caused by the vibration of the diaphragm is not smooth during operation. Under the condition of high voltage, the vibration amplitude of the diaphragm is large, and the gas in the cavity is turbulent, causing "sanding" and “hissing" airflow noise at certain frequencies, which seriously affects the user experience.
  • the existing detection methods for airflow noise include: (1) The micro speakers of smart devices play audio signals, which are judged and detected by human ears; (2) The audio signals emitted by the micro speakers of smart devices using electro-acoustic instruments Perform analysis and testing. These detection methods have high process cost, long cycle, difficult to guarantee accuracy, and limited versatility. In view of this, there is an urgent need to provide a new airflow noise detection method.
  • the present application provides an airflow noise detection method, device, computer equipment, and storage medium, which are used to solve the problem of low airflow noise elimination detection efficiency in the prior art.
  • an embodiment of the present application provides a method for detecting airflow noise, which is applied to a micro speaker, and the method includes:
  • a machine learning classifier is used to detect whether there is airflow noise in the original audio signal according to the audio feature.
  • an embodiment of the present application also provides an airflow noise detection device, the device including:
  • the signal collection module is used to collect the original audio signal in the micro speaker
  • the feature extraction module is used to perform feature extraction on the original audio signal to obtain audio features
  • the noise detection module is used to detect whether there is airflow noise in the original audio signal through a machine learning classifier and according to the audio characteristics.
  • the embodiments of the present application also provide a computer device, including a memory, a processor, and a computer program stored on the memory and running on the processor.
  • a computer program stored on the memory and running on the processor.
  • an embodiment of the present application also provides a computer-readable storage medium, including computer instructions, which when run on a computer, cause the computer to execute the steps of the airflow noise detection method described above.
  • the original audio signal in the micro speaker is collected, and then the original audio signal is feature extracted to obtain the audio feature, and finally through the machine learning classifier, and according to the audio feature Detect whether there is airflow noise in the original audio signal.
  • the machine learning classifier can reflect the inherent characteristics of airflow noise or the absence of airflow noise through training, the machine learning classifier can be used to detect whether there is airflow noise in the original audio signal, which is significantly improved
  • the accuracy of airflow noise detection is improved, and the cost is reduced. Since real-time detection or offline detection can be realized, the applicability of the detection method is improved.
  • Fig. 1 is a flow chart of the method for detecting airflow noise in an embodiment
  • Figure 2 is a flowchart of the audio feature extraction method in an embodiment
  • Figure 3 is a flowchart of the method for detecting airflow noise in another embodiment
  • Fig. 4 is a flowchart of the machine learning classifier training method in an embodiment
  • Fig. 5 is a flowchart of the method for acquiring the sample audio signal set in an embodiment
  • FIG. 6 is a schematic diagram of the structure of the airflow noise detection method device in an embodiment
  • FIG. 7 is a schematic diagram of the internal structure of a computer device running the above-mentioned airflow noise detection method in an embodiment.
  • a method for detecting airflow noise is proposed.
  • the realization of the method can rely on a computer program, which can run on a computer system based on the von Neumann system.
  • the airflow noise detection method provided in this embodiment is applied to a micro speaker, and the airflow noise detection method specifically includes the following steps:
  • Step 102 Collect the original audio signal in the micro speaker.
  • the original audio signal refers to the unprocessed signal in the WeChat speaker.
  • the sound information on the micro speaker can be collected through the sound card as the original audio signal, and the video or audio data can also be collected from the micro speaker as the original audio signal.
  • the original audio signal is due to the small cavity of the micro speaker and the precise structure, but the large amplitude of the diaphragm makes it easy for the airflow to form turbulence in the cavity and generate flow-induced noise. Therefore, in this embodiment, the original audio signal in the micro speaker is obtained. Audio signal for subsequent air noise detection based on the original audio signal.
  • Step 104 Perform feature extraction on the original audio signal to obtain audio features.
  • audio features are features used to characterize audio signals
  • the audio features can be audio features including time domain features, frequency domain features, and cepstrum domain features, or a combination of these features.
  • the time domain features, frequency domain features, and cepstral domain features can be Mel Cepstrum Coefficient (MFCC), linear prediction Common coefficient (LPCC) or component envelope, etc.
  • MFCC Mel Cepstrum Coefficient
  • LPCC linear prediction Common coefficient
  • a combination of time domain features, frequency domain characteristics, and cepstral domain characteristics can be used as audio features to improve the analysis of airflow noise in the original audio signal. Get accuracy.
  • Step 106 Detect whether there is airflow noise in the original audio signal according to the audio feature through the machine learning classifier.
  • a machine learning (Machine Learning, ML for short) classifier is a machine learning algorithm model with classification capabilities after training.
  • the machine learning classifier can have classification capabilities through sample learning.
  • the machine learning classifier of this embodiment is used to classify the original audio signal characterized by audio features into one of airflow noise signals and no airflow noise signals.
  • Machine learning classifiers can use SVM (Support Vector Machine (Support Vector Machine) classifier, logistic regression, ensemble method, random forest, neural network model, etc.
  • the ensemble method can be Boosting algorithm and Bagging algorithm and their variants.
  • the recognition accuracy of SVM classifier is more than 98% when detecting airflow noise, and when the integrated method is used, the recognition accuracy is more than 97%.
  • At least one machine learning model can be used to classify a classifier.
  • the training input is the audio features of various original audio signals that have been obtained, and the correspondence between the audio features and whether there is airflow noise is established.
  • the relationship classifier This makes the machine learning classifier have the ability to determine whether the input audio features have airflow noise.
  • the machine learning classifier is a two-classifier, that is, two classification results are obtained, that is, there is airflow noise or no airflow noise. Since the machine learning classifier can reflect the inherent characteristics of the presence or absence of airflow noise through training, the machine learning classifier can be used to detect whether there is airflow noise in the original audio signal, which significantly improves the accuracy of airflow noise detection.
  • the machine learning classifier is used to detect whether there is airflow noise in the original audio signal according to the audio characteristics. Applicability, compared with traditional instrument testing, on the basis of ensuring accuracy, it saves testing costs.
  • the above-mentioned airflow noise detection method collects the original audio signal in the micro speaker, and then extracts the characteristics of the original audio signal to obtain the audio characteristics. Finally, the machine learning classifier is used to detect whether the original audio signal has airflow noise according to the audio characteristics.
  • the machine learning classifier can reflect the inherent characteristics of the presence or absence of airflow noise through training, so that the machine learning classifier can be used to detect whether there is airflow noise in the original audio signal, which significantly improves the accuracy of airflow noise detection and reduces the cost. , And improve the applicability of the detection method.
  • performing feature extraction on the original audio signal to obtain audio features includes:
  • Step 104A segment the original audio signal to obtain multiple signal segments
  • Step 104B Perform feature extraction on each signal segment separately to obtain sub audio features
  • Step 104C Combine the sub-audio features according to signal segments to obtain audio features.
  • the signal segment refers to a segment of the audio signal in the original audio signal.
  • the original audio signal can be divided into multiple audio frames, and the division method can be an overlap framing method, that is, the end of the previous frame of the original audio signal The part (such as 100 milliseconds) is used as the beginning part of the next frame of the original audio signal, and multiple signal segments of the original audio signal can be obtained after overlapping and framing.
  • the sub-audio features in this embodiment are the same as the audio features in step 104, and will not be repeated here.
  • the multiple sub-audio features are combined according to the sequence corresponding to the signal segment.
  • the combination method may be to combine all the sub-audio features, or select a specific number of sub-audio features to combine to obtain the audio feature. Understandably, by segmenting the original audio signal and then performing feature extraction, the features of each signal segment are made more uniform and accurate, so as to facilitate subsequent more reliable analysis.
  • feature extraction can speed up the feature extraction, so that the features of the signal segment are more unified and rich, thereby improving the accuracy of audio features.
  • the method before the feature extraction is performed on each signal segment separately to obtain the sub-audio feature, the method further includes:
  • each sub-audio feature corresponding to each signal segment will be relatively wide.
  • These signal segments can be adjusted to a preset interval through normalization processing, so that the features of each signal segment are more obvious and rich, thereby further ensuring the performance of each signal segment. Unification helps to improve the accuracy of the sub-audio features corresponding to the signal segment.
  • the method further includes:
  • Step 108 Obtain sample audio signals and corresponding marks in the sample audio signal set, where the mark of the positive sample audio signal is no airflow noise, and the mark of the negative sample audio signal is airflow noise;
  • Step 110 Extract the audio features of each sample in the sample audio signal set to obtain a sample audio feature set
  • Step 112 Train a machine learning classifier according to the sample audio feature set and corresponding labels.
  • the original audio signal played by the WeChat speaker One part of the original audio signal contains obvious airflow noise, and the other part contains no airflow noise.
  • the original audio signal is divided into signal segments for labeling, and then the characteristics of each audio signal are extracted.
  • Features include time domain features, frequency domain features, cepstrum domain features, etc.
  • the audio features of the audio signal of the positive sample are mixed with the audio feature of the audio signal of the negative sample.
  • the mixing can be random mixing.
  • the corresponding labels constitute a sample audio feature set.
  • the sample audio signal set includes positive samples and negative samples, and the negative samples include the characteristics of audio features with airflow noise.
  • the classifier can learn more accurate classification rules, which can further improve the accuracy of airflow noise detection.
  • the audio feature includes at least one of a time domain feature, a frequency domain feature, and a cepstrum domain feature.
  • the time domain features include short-term average zero-crossing rate and short-term autocorrelation function; frequency domain features include extracting short-term power spectral density functions; cepstrum domain features Mel frequency cepstral coefficients and linear prediction cepstrum coefficients.
  • at least one audio feature is extracted to perform audio signal analysis using the characteristics of different audio features to improve the accuracy of detection.
  • time domain feature, frequency domain feature, and cepstrum domain feature or a combination of these features can be trained by a machine learning classifier, and then the optimal audio feature can be determined according to the machine learning classifier .
  • training a machine learning classifier according to the sample audio feature set and corresponding labels includes:
  • Step 112A Obtain a set of discrete parameter values of the machine learning classifier
  • Step 112B According to the value of each parameter in the discrete parameter value set and the sample audio feature set, train a machine learning classifier corresponding to each parameter value, and obtain the classification of the machine learning classifier corresponding to the corresponding parameter value Prediction accuracy
  • Step 112C Screen the maximum classification prediction accuracy rate and obtain corresponding parameter values and audio features of the sample audio feature set, and train a machine learning classifier according to the obtained parameter values and sample audio feature set.
  • the discrete parameter value set is a collection of several discrete parameter values.
  • the parameter value is the value of the parameter required by the training machine to learn the classifier. Specifically, according to the first step, sampling in the continuous parameter value range to obtain a series of discrete parameter values to form a discrete parameter value set. If the machine learning classifier includes multiple parameters that need to be learned, a set of discrete parameter values corresponding to each parameter can be obtained. If the machine learning classifier adopts the SVM classifier, the parameters such as the penalty coefficient.
  • each parameter value in the discrete parameter value set can be traversed, and the current traversed parameter value and the audio feature corresponding to the sample audio feature set can be used to train the machine learning classifier, and obtain the machine learning classifier corresponding The classification prediction accuracy rate until all the parameter values in the discrete parameter value set and the audio features corresponding to the sample audio feature set are traversed.
  • the sample audio feature set is divided into a training set and a test set, and each parameter value in the discrete parameter value set and the audio feature corresponding to the sample audio feature set are traversed, and the current traversed parameter values and the samples in the training set are used
  • the audio feature corresponding to the audio feature set trains the machine learning classifier, and uses the trained machine learning classifier to predict the test set, obtains the known classification results of the test set, and compares the predicted prediction results with the known classification results. Obtain the correct rate of classification prediction of the corresponding machine learning classifier.
  • the sample audio feature set is used to quickly find suitable parameter values and audio features, so that the parameter values, audio features, and sample audio feature sets are used for training, which can improve the efficiency of training machine learning classifiers.
  • acquiring sample audio signals and corresponding marks in a sample audio signal set includes:
  • Step 108A In the same environment, using the same audio collection device, collecting the sample audio signal set by adjusting the gain of the micro speaker;
  • Step 108B segment the sample audio signals in the sample audio signal set to obtain multiple sample audio segments corresponding to each sample audio signal;
  • Step 108C According to the sample audio segment, the label of the sample audio signal is determined through the preset speaker model.
  • the same audio collection equipment is used to collect the sample audio signal set by adjusting the gain of the micro speaker, that is, the high point of the voltage is controlled by adjusting the gain, thereby controlling the vibration amplitude of the diaphragm, thereby ensuring that the collected original audio signal is rich It improves the richness and comprehensiveness of the sample audio signal set.
  • the preset speaker model determines the mark of the sample audio signal, that is, the accuracy of the mark of the sample audio signal is ensured by the method of continuing learning. In this embodiment, by acquiring a comprehensive and rich sample audio signal set and performing accurate labeling, it is beneficial to improve the accuracy of machine learning classifier training.
  • an embodiment of the present application provides an airflow noise detection device 600, as shown in FIG. 6, including: a signal collection module 602 for collecting the original audio signal in a micro speaker; a feature extraction module 604 for checking Feature extraction is performed on the original audio signal to obtain audio features; the noise detection module 606 is configured to use a machine learning classifier to detect whether there is airflow noise in the original audio signal according to the audio features.
  • the airflow noise detection device 600 of this embodiment includes: a signal collection module 602, used to collect the original audio signal in the micro speaker; a feature extraction module 604, used to compare the original audio signal Perform feature extraction to obtain audio features; the noise detection module 606 is configured to use a machine learning classifier to detect whether there is airflow noise in the original audio signal according to the audio features. Since the machine learning classifier can reflect the inherent characteristics of the presence or absence of airflow noise through training, the machine learning classifier can be used to detect whether there is airflow noise in the original audio signal, which significantly improves the accuracy of airflow noise detection and reduces Cost, and improve the applicability of the detection method.
  • Fig. 7 shows an internal structure diagram of a computer device in an embodiment.
  • the computer device may specifically be a server or a terminal.
  • the computer device 700 includes a processor 710, a memory 720, and a network interface 730 connected through a system bus.
  • the memory 720 includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium of the computer device stores an operating system and may also store a computer program.
  • the processor can realize the method for detecting airflow noise.
  • a computer program can also be stored in the internal memory, and when the computer program is executed by the processor, the processor can execute the method for detecting airflow noise.
  • FIG. 7 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer components than shown in FIG. 7, or combining some components, or having a different component arrangement.
  • the method for detecting airflow noise provided by the present application can be implemented in the form of a computer program, and the computer program can be run on a computer device as shown in FIG. 7.
  • the memory of the computer device can store various program modules that make up the device for detecting airflow noise. For example, the signal acquisition module 602, the feature extraction module 604, and the noise detection module 606.
  • a computer device includes a memory and a processor.
  • the memory stores a computer program.
  • the processor executes the following steps: collecting the original audio signal in a micro speaker; Perform feature extraction on the original audio signal to obtain audio features; use a machine learning classifier to detect whether there is airflow noise in the original audio signal according to the audio features.
  • performing feature extraction on the original audio signal to obtain audio features includes: segmenting the original audio signal to obtain multiple signal segments; performing feature extraction on each of the signal segments to obtain Sub-audio features; combining the sub-audio features according to the signal segment to obtain the audio feature.
  • the method before the feature extraction is performed on each of the signal segments to obtain sub-audio features, the method further includes: normalizing the signal segments.
  • the method further includes: obtaining sample audio signals and corresponding marks in the sample audio signal set, wherein the mark of the positive sample audio signal is no airflow noise, and the mark of the negative sample audio signal is airflow noise; Extract each sample audio feature in the sample audio signal set to obtain a sample audio feature set; train a machine learning classifier according to the sample audio feature set and corresponding tags.
  • the audio feature includes at least one of a time domain feature, a frequency domain feature, and a cepstrum domain feature.
  • the training of a machine learning classifier according to the sample audio feature set and corresponding labels includes: obtaining a set of discrete parameter values of the machine learning classifier; For each parameter value and the sample audio feature set, train a machine learning classifier corresponding to each parameter value, and obtain the classification prediction accuracy rate of the machine learning classifier corresponding to the corresponding parameter value; filter out the largest classification Predict the correct rate and obtain the corresponding parameter values and audio features of the sample audio feature set, and train the machine learning classifier according to the obtained parameter values and the sample audio feature set.
  • the acquiring the sample audio signals and the corresponding marks in the sample audio signal set includes: in the same environment, using the same audio collection device, and collecting the samples by adjusting the gain of the micro-speaker Audio signal set; segment the sample audio signals in the sample audio signal set to obtain the multiple sample audio segments corresponding to each sample audio signal; according to the sample audio segments, the preset speaker model is used to determine the The mark of the sample audio signal.
  • a computer-readable storage medium wherein the computer-readable storage medium stores a computer program, and is characterized in that, when the computer program is executed by a processor, the following steps are implemented: collecting the original audio signal in a micro speaker; Feature extraction is performed on the audio signal to obtain audio features; a machine learning classifier is used to detect whether there is airflow noise in the original audio signal according to the audio features.
  • performing feature extraction on the original audio signal to obtain audio features includes: segmenting the original audio signal to obtain multiple signal segments; performing feature extraction on each of the signal segments to obtain Sub-audio features; combining the sub-audio features according to the signal segment to obtain the audio feature.
  • the method before the feature extraction is performed on each of the signal segments to obtain sub-audio features, the method further includes: normalizing the signal segments.
  • the method further includes: obtaining sample audio signals and corresponding marks in the sample audio signal set, wherein the mark of the positive sample audio signal is no airflow noise, and the mark of the negative sample audio signal is airflow noise; Extract each sample audio feature in the sample audio signal set to obtain a sample audio feature set; train a machine learning classifier according to the sample audio feature set and corresponding tags.
  • the audio feature includes at least one of a time domain feature, a frequency domain feature, and a cepstrum domain feature.
  • the training of a machine learning classifier according to the sample audio feature set and corresponding labels includes: obtaining a set of discrete parameter values of the machine learning classifier; For each parameter value and the sample audio feature set, train a machine learning classifier corresponding to each parameter value, and obtain the classification prediction accuracy rate of the machine learning classifier corresponding to the corresponding parameter value; filter out the largest classification Predict the correct rate and obtain the corresponding parameter values and audio features of the sample audio feature set, and train the machine learning classifier according to the obtained parameter values and the sample audio feature set.
  • the acquiring the sample audio signals and the corresponding marks in the sample audio signal set includes: in the same environment, using the same audio collection device, and collecting the samples by adjusting the gain of the micro-speaker Audio signal set; segment the sample audio signals in the sample audio signal set to obtain the multiple sample audio segments corresponding to each sample audio signal; according to the sample audio segments, the preset speaker model is used to determine the The mark of the sample audio signal.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Channel
  • memory bus Radbus direct RAM
  • RDRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Circuit For Audible Band Transducer (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

本申请实施例公开了一种气流杂音检测方法，通过采集微型扬声器中的原始音频信号，然后对原始音频信号进行特征提取，得到音频特征，最后通过机器学习分类器，并根据音频特征检测原始音频信号是否存在气流杂音，由于机器学习分类器就可以通过训练反映出存在气流杂音或者不存在气流杂音的内在特性，从而利用该机器学习分类器检测原始音频信号是否存在气流杂音，显著提高了气流杂音检测准确率，降低了成本，并提高了检测方法的适用性。此外，还提出一种气流杂音检测装置、计算机设备及存储介质。

Description

气流杂音检测方法、装置、终端及存储介质 技术领域
本申请涉及计算机技术领域，尤其涉及一种气流杂音检测方法、装置、终端及存储介质。
 背景技术
微型扬声器腔体狭小，工作时振膜振动造成的气流流动并不顺畅。在高电压条件下，振膜运动幅度较大，腔体内气体发生湍流，在某些频率下导致“沙沙”、“嘶嘶”的气流噪声，这严重影响用户体验。
 技术问题
随着智能设备如手机、平板等的功放应用的推广，大电压下播放音乐时的气流噪声问题显得日益严重。为了改善这一问题，需要检测微型扬声器是否存在杂音。目前,现有的气流杂音的检测方法有：（1）、智能设备的微型扬声器播放音频信号，通过人耳去判断检测； (2)、采用电声仪器对智能设备的微型扬声器发出的音频信号进行分析检测。这些检测方法工艺成本较高、周期较长，准确性难以保证，且通用性受到限制，鉴于此，亟需提供一种新的气流杂音检测方法。
 技术解决方案
有鉴于此，本申请提供了一种气流杂音检测方法、装置、计算机设备及存储介质，用于解决现有技术中气流杂音消除检测效率不高的问题。
本申请实施例的具体技术方案为：
第一方面，本申请实施例提供一种气流杂音检测方法，应用于微型扬声器，所述方法包括：
采集微型扬声器中的原始音频信号；
对所述原始音频信号进行特征提取，得到音频特征；
通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
第二方面，本申请实施例还提供一种气流杂音检测装置，所述装置包括：
信号采集模块，用于采集微型扬声器中的原始音频信号；
特征提取模块，用于对所述原始音频信号进行特征提取，得到音频特征；
杂音检测模块，用于通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
第三方面，本申请实施例还提供一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上所述气流杂音检测方法的步骤。
第四方面，本申请实施例还提供一种计算机可读存储介质，包括计算机指令，当所述计算机指令在计算机上运行时，使得计算机执行如上所述气流杂音检测方法的步骤。
 有益效果
实施本申请实施例，将具有如下有益效果：
采用了上述气流杂音检测方法、装置、终端及存储介质之后，通过采集微型扬声器中的原始音频信号，然后对原始音频信号进行特征提取，得到音频特征，最后通过机器学习分类器，并根据音频特征检测原始音频信号是否存在气流杂音，由于机器学习分类器就可以通过训练反映出存在气流杂音或者不存在气流杂音的内在特性，从而利用该机器学习分类器检测原始音频信号是否存在气流杂音，显著提高了气流杂音检测准确率，降低了成本，由于能够实现实时检测或者离线检测，并提高了检测方法的适用性。
 附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
其中：
图1为一个实施例中所述气流杂音检测方法的流程图；
图2为一个实施例中所述音频特征提取方法的流程图；
图3为另一个实施例中所述气流杂音检测方法的流程图；
图4为一个实施例中所述机器学习分类器训练方法的流程图；
图5为一个实施例中所述样本音频信号集获取方法的流程图；
图6为一个实施例中所述气流杂音检测方法装置的结构示意图；
图7为一个实施例中运行上述气流杂音检测方法的计算机设备的内部结构示意图。
 本发明的实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
为解决传统技术中由于人工检测或者仪器检测的准确性不高且成本较高的问题。
基于上述问题，在本实施例中，特提出了一种气流杂音检测方法。该方法的实现可依赖于计算机程序，该计算机程序可运行于基于冯诺依曼体系的计算机系统之上。
如图1所示，本实施例提供的气流杂音检测方法，应用于微型扬声器，该气流杂音检测方法具体包括以下步骤：
步骤102：采集微型扬声器中的原始音频信号。
其中，原始音频信号是指微信扬声器中的未经处理的信号。具体地，可以通过声卡采集微型扬声器上的声音信息作为原始音频信号，也可以从微型扬声器中采集视频或者音频数据作为原始音频信号。通常，原始音频信号由于微型扬声器腔体较小，结构精密，但是振膜振幅较大，使得气流在腔体内容易形成湍流并产生流致噪声，因此，本实施例中通过获取微型扬声器中的原始音频信号，以便后续基于该原始音频信号进行气流杂音检测。
步骤104：对原始音频信号进行特征提取，得到音频特征。
其中，音频特征是用于表征音频信号的特征，其中的音频特征可以是音频特征包括时域特征、频域特征及倒谱域特征或者这几种特征的组合。具体地，可以通过C/C++ 、Python、MATLAB等工具包对原始音频信号进行特征提取，得到音频特征，更具体地，其中的时域特征、频域特征及倒谱域特征可以是梅尔倒谱系数（MFCC）、线性预测普系数（LPCC）或者分量包络等。作为本实施例的优选，为了更加准确地分辨气流杂音，可以利用时域特征、频域特征及倒谱域特征这几种特征的组合作为音频特征，以便提高对原始音频信号中的气流杂音分析得准确度。
步骤106：通过机器学习分类器，并根据音频特征检测原始音频信号是否存在气流杂音。
其中，机器学习（Machine Learning，简称ML）分类器是经过训练后具有分类能力的机器学习算法模型。机器学习分类器可通过样本学习具备分类能力， 本实施例的机器学习分类器用于将由音频特征表征的原始音频信号划分到有气流杂音信号和没有气流杂音信号中的一类。机器学习分类器可以采用SVM(Support Vector Machine，支持向量机)分类器、逻辑回归、集成方法、随机森林、神经网络模型等，其中的集成方法又可以是Boosting算法和Bagging算法及其变种。实践中采用SVM分类器在检测气流噪声时，识别精度达到98%以上，采用集成方法在检测气流噪声时，识别精度达到97%以上。
具体地，可以利用至少一个机器学习模型进行分类的分类器，作为机器学习分类器训练的部分，训练输入是各种已经获得的原始音频信号的音频特征，建立音频特征与是否存在气流杂音的对应的关系分类器。使得该机器学习分类器具备判断输入的音频特征对否存在气流杂音的能力。本实施例中，该机器学习分类器为二分类器，即得到两个分类结果，也即存在气流杂音或者不存在气流杂音。由于机器学习分类器就可以通过训练反映出存在气流杂音或者不存在气流杂音的内在特性，从而利用该机器学习分类器检测原始音频信号是否存在气流杂音，显著提高了气流杂音检测准确率。
值得说明的是，本实施例中，通过机器学习分类器，并根据音频特征检测原始音频信号是否存在气流杂音，能够适用于离线检测或者实时检测的气流杂音检测场景，提高了气流杂音检测方法的适用性，相较于传统的仪器检测，在保证准确度的基础上，节省了检测成本。
上述气流杂音检测方法，通过采集微型扬声器中的原始音频信号，然后对原始音频信号进行特征提取，得到音频特征，最后通过机器学习分类器，并根据音频特征检测原始音频信号是否存在气流杂音，由于机器学习分类器就可以通过训练反映出存在气流杂音或者不存在气流杂音的内在特性，从而利用该机器学习分类器检测原始音频信号是否存在气流杂音，显著提高了气流杂音检测准确率，降低了成本，并提高了检测方法的适用性。
如图2所示，在一个实施例中，对原始音频信号进行特征提取，得到音频特征，包括：
步骤104A：对原始音频信号进行分割，得到多个信号段；
步骤104B：分别对每个信号段进行特征提取，得到子音频特征；
步骤104C：将子音频特征按照信号段进行组合，得到音频特征。
其中，信号段是指原始音频信号中的一段音频信号，具体地，可以将原始音频信号划分为多个音频帧，划分的方式可以采用重叠分帧方式，即将原始音频信号的前一帧的末尾部分（如100毫秒)作为原始音频信号的后一帧的起始部分，经过重叠分帧，可以得到原始音频信号的多个信号段。还可以通过MATLAB工具中的分帧函数如enframe()，对原始音频信号进行分帧，得到原始音频信号的多个信号段。然后对划分出来的每个信号段分别进行特征提取，得到每个信号段对应的子音频特征，本实施例中的子音频特征与步骤104中的音频特征相同，此处不在赘述。最后，将多个子音频特征分别按照信号段对应的顺序进行组合，其中的组合方式可以是将所有子音频特征进行组合，也可以是选取特定数量的子音频特征进行组合，得到音频特征。可以理解地，通过对原始音频信号进行分割后再进行特征提取，使得每个信号段的特征更加统一和准确，以便于后续进行更加可靠地分析。
进一步地，对应离线采集不同时段的原始音频信号，通过分割后，再进行特征提取能够加快特征提取的速度，使得信号段的特征更加统一丰富，进而提高了音频特征的准确性。
在一个实施例中，在分别对每个信号段进行特征提取，得到子音频特征之前，还包括：
对信号段进行归一化处理。
其中，各信号段对应的各个子音频特征分布会比较广，可以通过归一化处理将这些信号段调整到预设区间，使得各个信号段的特征更加明显丰富，从而进一步保证了各个信号段的统一，有利于提高信号段对应的子音频特征的准确度。
如图3所示，在一个实施例中，该方法还包括：
步骤108：获取样本音频信号集中的样本音频信号和相应的标记，其中，正样本音频信号的标记为无气流杂音，负样本音频信号的标记为有气流杂音；
步骤110：提取样本音频信号集中各个样本音频特征，得到样本音频特征集；
步骤112：根据样本音频特征集和相应的标记，训练机器学习分类器。
具体地，采集微信扬声器播放的原始音频信号，其中一部分原始音频信号包含明显的气流噪声，另一部分则没有气流噪声，将始音频信号分割为信号段进行标注，然后提取各个音频信号的特征，该特征包括时域特征、频域特征、倒谱域特征等，将正样本的音频信号的音频特征与负样本的音频信号的音频特征进行混合，其中的混合可采用随机混合，所有样本音频特征与对应的标记构成样本音频特征集，本实施例中，样本音频信号集中包括正样本和负样本，负样本则包括存在气流杂音的音频特征的特性，利用这样的样本音频信号集训练出的机器学习分类器能够学习到更加准确的分类规则，从而可以进一步提高气流杂音检测的正确率。
在一个实施例中，音频特征包括时域特征、频域特征及倒谱域特征中的至少一种。
其中，时域特征包括短时平均过零率和短时自相关函数；频域特征包括提取短时功率谱密度函数；倒谱域特征梅尔频率倒谱系数和线性预测倒谱系数。本实施例中，通过提取至少一种音频特征，以利用不同音频特征的特点进行音频信号分析，提高检测的准确度。
值得说明的是，本实施例中，可以通过机器学习分类器对该时域特征、频域特征及倒谱域特征或者这些特征的组合进行训练后，根据机器学习分类器确定最优的音频特征。
如图4所示，在一个实施例中，根据样本音频特征集和相应的标记，训练机器学习分类器，包括：
步骤112A：获取机器学习分类器的离散参数取值集合；
步骤112B：根据离散参数取值集合中的每个参数取值、样本音频特征集，训练与每个参数取值相应的机器学习分类器，并获得相应参数取值所对应机器学习分类器的分类预测正确率；
步骤112C：筛选出最大的分类预测正确率并获取相应的参数取值和样本音频特征集的音频特征，并根据获取的参数取值和样本音频特征集训练机器学习分类器。
其中，离散参数取值集合是若干离散的参数取值构成的集合。参数取值是训练机 器学习分类器所需参数的取值。具体可按照第一步长，在连续参数取值范围中采样，获得一系列的离散参数取值，以构成离散参数取值集合。若机器学习分类器包括多个需要学习的参数，则可获取与每个参数对应的离散参数取值集 合。若机器学习分类器采用SVM分类器，则参数比如惩罚系数。具体地，可以遍历离散参数取值集合中的每个参数取值，分别利用当前遍历的参数取值、样本音频特征集对应的音频特征训练机器学习分类器，并获得该机器学习分类器对应的分类预测正确率，直至遍历完离散参数取值集合中的所有参数取值和样本音频特征集对应的音频特征。
进一步地，将样本音频特征集划分为训练集和测试集，遍历离散参数取值集合中的每个参数取值、样本音频特征集对应的音频特征，利用当前遍历的参数取值、训练集中样本音频特征集对应的音频特征训练机器学习分类器，并利用训练的机器学习分类器对测试集进行预测，获取测试集已知的分类结果，将预测得到的预测结果与已知的分类结果比较，得到相应机器学习分类器的分类预测正确率。将步骤112B中获得的分类预测正确率进行比较，找出其中最大的分类预测正确率，获取训练该最大的分类预测正确率的机器学习分类器所用的参数取值和样本音频特征集对应的音频特征，从而利用获取的参数取值、样本音频特征集中对应的音频特征继续训练机器学习分类器。
本实施例中，利用样本音频特征集快速找出合适的参数取值以及音频特征，从而利用该参数取值、音频特征以及样本音频特征集进行训练，可提高训练机器学习分类器的效率。
如图5所示，在一个实施例中，获取样本音频信号集中的样本音频信号和相应的标记，包括：
步骤108A：在同一的环境下，采用相同的音频采集设备，通过调节微型扬声器的增益采集得到样本音频信号集；
步骤108B：对样本音频信号集中的样本音频信号进行分割，得到每个样本音频信号对应的多个样本音频段；
步骤108C：根据样本音频段，通过预设的扬声器模型确定样本音频信号的标记。
具体地，为了保证采集的原始音频信号多样性和带表性，保证其中一部分原始音频信号包含气流噪声，另一部分没有气流噪声，因此，同一的环境下，避免硬件和人为干扰。同时采用相同的音频采集设备，通过调节微型扬声器的增益采集得到样本音频信号集，即通过调节增益来控制电压的高点，从而控制振膜振动幅度，从而保证了采集到的的原始音频信号丰富性，提高了样本音频信号集的丰富全面性。
对样本音频信号集中的样本音频信号进行分割，得到每个样本音频信号对应的多个样本音频段，通过对样本音频信号进行分割，有利于提高样本音频段的统一性，根据样本音频段，通过预设的扬声器模型确定样本音频信号的标记，即通过继续学习的方法保证样本音频信号标记的准确性。本实施例中，通过获取全面丰富的样本音频信号集，并进行准确标记，有利于提高机器学习分类器训练的准确性。
基于同一申请构思，本申请实施例提供一种气流杂音检测装置600，如图6所示，包括：信号采集模块602，用于采集微型扬声器中的原始音频信号；特征提取模块604，用于对所述原始音频信号进行特征提取，得到音频特征；杂音检测模块606，用于通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
具体地，本实施例的气流杂音检测装置600，如图6所示，包括：信号采集模块602，用于采集微型扬声器中的原始音频信号；特征提取模块604，用于对所述原始音频信号进行特征提取，得到音频特征；杂音检测模块606，用于通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。由于机器学习分类器就可以通过训练反映出存在气流杂音或者不存在气流杂音的内在特性，从而利用该机器学习分类器检测原始音频信号是否存在气流杂音，显著提高了气流杂音检测准确率，降低了成本，并提高了检测方法的适用性。
需要说明的是，本实施例中气流杂音检测的装置的实现与上述气流杂音检测的方法的实现思想一致，其实现原理在此不再进行赘述，可具体参阅上述方法中对应内容。
图7示出了一个实施例中计算机设备的内部结构图。该计算机设备具体可以是服务器，也可以是终端。如图7所示，该计算机设备700包括通过系统总线连接的处理器710、存储器720和网络接口730。其中，存储器720包括非易失性存储介质和内存储器。该计算机设备的非易失性存储介质存储有操作系统，还可存储有计算机程序，该计算机程序被处理器执行时，可使得处理器实现气流杂音检测的方法。该内存储器中也可储存有计算机程序，该计算机程序被处理器执行时，可使得处理器执行气流杂音检测的方法。本领域技术人员可以理解，图7中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图7中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，本申请提供的气流杂音检测的方法可以实现为一种计算机程序的形式，计算机程序可在如图7所示的计算机设备上运行。计算机设备的存储器中可存储组成所述气流杂音检测的装置的各个程序模块。比如，信号采集模块602，特征提取模块604，杂音检测模块606。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述计算机程序被所述处理器执行时，使得所述处理器执行以下步骤：采集微型扬声器中的原始音频信号；对所述原始音频信号进行特征提取，得到音频特征；通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
在一个实施例中，对所述原始音频信号进行特征提取，得到音频特征，包括：对所述原始音频信号进行分割，得到多个信号段；分别对每个所述信号段进行特征提取，得到子音频特征；将所述子音频特征按照所述信号段进行组合，得到所述音频特征。
在一个实施例中，在所述分别对每个所述信号段进行特征提取，得到子音频特征之前，还包括：对所述信号段进行归一化处理。
在一个实施例中，所述方法还包括：获取样本音频信号集中的样本音频信号和相应的标记，其中，正样本音频信号的标记为无气流杂音，负样本音频信号的标记为有气流杂音；提取所述样本音频信号集中各个样本音频特征，得到样本音频特征集；根据所述样本音频特征集和相应的标记，训练机器学习分类器。
在一个实施例中，所述音频特征包括时域特征、频域特征及倒谱域特征中的至少一种。
在一个实施例中，所述根据所述样本音频特征集和相应的标记，训练机器学习分类器，包括：获取机器学习分类器的离散参数取值集合；根据所述离散参数取值集合中的每个参数取值、所述样本音频特征集，训练与每个参数取值相应的机器学习分类器，并获得相应参数取值所对应机器学习分类器的分类预测正确率；筛选出最大的分类预测正确率并获取相应的参数取值和样本音频特征集的音频特征，并根据获取的参数取值和所述样本音频特征集训练机器学习分类器。
在一个实施例中，所述获取样本音频信号集中的样本音频信号和相应的标记，包括：在同一的环境下，采用相同的音频采集设备，通过调节所述微型扬声器的增益采集得到所述样本音频信号集；对所述样本音频信号集中的样本音频信号进行分割，得到每个样本音频信号对应的所述多个样本音频段；根据所述样本音频段，通过预设的扬声器模型确定所述样本音频信号的标记。
一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如下步骤：采集微型扬声器中的原始音频信号；对所述原始音频信号进行特征提取，得到音频特征；通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
在一个实施例中，对所述原始音频信号进行特征提取，得到音频特征，包括：对所述原始音频信号进行分割，得到多个信号段；分别对每个所述信号段进行特征提取，得到子音频特征；将所述子音频特征按照所述信号段进行组合，得到所述音频特征。
在一个实施例中，在所述分别对每个所述信号段进行特征提取，得到子音频特征之前，还包括：对所述信号段进行归一化处理。
在一个实施例中，所述方法还包括：获取样本音频信号集中的样本音频信号和相应的标记，其中，正样本音频信号的标记为无气流杂音，负样本音频信号的标记为有气流杂音；提取所述样本音频信号集中各个样本音频特征，得到样本音频特征集；根据所述样本音频特征集和相应的标记，训练机器学习分类器。
在一个实施例中，所述音频特征包括时域特征、频域特征及倒谱域特征中的至少一种。
在一个实施例中，所述根据所述样本音频特征集和相应的标记，训练机器学习分类器，包括：获取机器学习分类器的离散参数取值集合；根据所述离散参数取值集合中的每个参数取值、所述样本音频特征集，训练与每个参数取值相应的机器学习分类器，并获得相应参数取值所对应机器学习分类器的分类预测正确率；筛选出最大的分类预测正确率并获取相应的参数取值和样本音频特征集的音频特征，并根据获取的参数取值和所述样本音频特征集训练机器学习分类器。
在一个实施例中，所述获取样本音频信号集中的样本音频信号和相应的标记，包括：在同一的环境下，采用相同的音频采集设备，通过调节所述微型扬声器的增益采集得到所述样本音频信号集；对所述样本音频信号集中的样本音频信号进行分割，得到每个样本音频信号对应的所述多个样本音频段；根据所述样本音频段，通过预设的扬声器模型确定所述样本音频信号的标记。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一非易失性计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上所揭露的仅为本申请较佳实施例而已，当然不能以此来限定本申请之权利范围，因此依本申请权利要求所作的等同变化，仍属本申请所涵盖的范围。

Claims (10)

  1. 一种气流杂音检测方法，其特征在于，应用于微型扬声器，所述方法包括：
    采集微型扬声器中的原始音频信号；
    对所述原始音频信号进行特征提取，得到音频特征；
    通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
  2. 如权利要求1所述气流杂音检测方法，其特征在于，所述对所述原始音频信号进行特征提取，得到音频特征，包括：
    对所述原始音频信号进行分割，得到多个信号段；
    分别对每个所述信号段进行特征提取，得到子音频特征；
    将所述子音频特征按照所述信号段进行组合，得到所述音频特征。
  3. 如权利要求2所述气流杂音检测方法，其特征在于，在所述分别对每个所述信号段进行特征提取，得到子音频特征之前，还包括：
    对所述信号段进行归一化处理。
  4. 如权利要求1所述气流杂音检测方法，其特征在于，所述方法还包括：
    获取样本音频信号集中的样本音频信号和相应的标记，其中，正样本音频信号的标记为无气流杂音，负样本音频信号的标记为有气流杂音；
    提取所述样本音频信号集中各个样本音频特征，得到样本音频特征集；
    根据所述样本音频特征集和相应的标记，训练机器学习分类器。
  5. 如权利要求1所述气流杂音检测方法，其特征在于，所述音频特征包括时域特征、频域特征及倒谱域特征中的至少一种。
  6. 如权利要求5所述气流杂音检测方法，其特征在于，所述根据所述样本音频特征集和相应的标记，训练机器学习分类器，包括：
    获取机器学习分类器的离散参数取值集合；
    根据所述离散参数取值集合中的每个参数取值、所述样本音频特征集，训练与每个参数取值相应的机器学习分类器，并获得相应参数取值所对应机器学习分类器的分类预测正确率；
    筛选出最大的分类预测正确率并获取相应的参数取值和样本音频特征集的音频特征，并根据获取的参数取值和所述样本音频特征集训练机器学习分类器。
  7. 如权利要求4所述气流杂音检测方法，其特征在于，所述获取样本音频信号集中的样本音频信号和相应的标记，包括：
    在同一的环境下，采用相同的音频采集设备，通过调节所述微型扬声器的增益采集得到所述样本音频信号集；
    对所述样本音频信号集中的样本音频信号进行分割，得到每个样本音频信号对应的多个样本音频段；
    根据所述样本音频段，通过预设的扬声器模型确定所述样本音频信号的标记。
  8. 一种气流杂音检测装置，其特征在于，所述装置包括：
    信号采集模块，用于采集微型扬声器中的原始音频信号；
    特征提取模块，用于对所述原始音频信号进行特征提取，得到音频特征；
    杂音检测模块，用于通过机器学习分类器，并根据所述音频特征检测所述原始音频信号是否存在气流杂音。
  9. 一种计算机设备，其特征在于，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至7中任一项所述气流杂音检测方法的步骤。
  10. 一种计算机可读存储介质，其特征在于，包括计算机指令，当所述计算机指令在计算机上运行时，使得计算机执行如权利要求1至7中任一项所述的气流杂音检测方法的步骤。
PCT/CN2020/096685 2020-06-12 2020-06-18 气流杂音检测方法、装置、终端及存储介质 WO2021248522A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010536859.XA CN111797708A (zh) 2020-06-12 2020-06-12 气流杂音检测方法、装置、终端及存储介质
CN202010536859.X 2020-06-12

Publications (1)

Publication Number Publication Date
WO2021248522A1 true WO2021248522A1 (zh) 2021-12-16

Family

ID=72803904

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/096685 WO2021248522A1 (zh) 2020-06-12 2020-06-18 气流杂音检测方法、装置、终端及存储介质

Country Status (2)

Country Link
CN (1) CN111797708A (zh)
WO (1) WO2021248522A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112487865A (zh) * 2020-11-02 2021-03-12 杭州兆华电子有限公司 一种基于机器学习的扬声器自动分类方法
CN112863548A (zh) * 2021-01-22 2021-05-28 北京百度网讯科技有限公司 训练音频检测模型的方法、音频检测方法及其装置
CN114627891A (zh) * 2022-05-16 2022-06-14 山东捷瑞信息技术产业研究院有限公司 一种动圈扬声器质量检测方法和装置
CN116038076B (zh) * 2023-03-28 2023-10-13 苏芯物联技术(南京)有限公司 一种基于场景匹配的焊接保护气体控制方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106202952A (zh) * 2016-07-19 2016-12-07 南京邮电大学 一种基于机器学习的帕金森疾病诊断方法
CN108073856A (zh) * 2016-11-14 2018-05-25 华为技术有限公司 噪音信号的识别方法及装置
CN109903752A (zh) * 2018-05-28 2019-06-18 华为技术有限公司 对齐语音的方法和装置
US20190297412A1 (en) * 2018-03-26 2019-09-26 Sony Corporation Apparatuses and Methods for Acoustic Noise Cancelling

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104078050A (zh) * 2013-03-26 2014-10-01 杜比实验室特许公司 用于音频分类和音频处理的设备和方法
CN103824557B (zh) * 2014-02-19 2016-06-15 清华大学 一种具有自定义功能的音频检测分类方法
CN108287816B (zh) * 2017-01-10 2021-06-04 腾讯科技（深圳）有限公司 兴趣点在线检测、机器学习分类器训练方法和装置
CN108764114B (zh) * 2018-05-23 2022-09-13 腾讯音乐娱乐科技（深圳）有限公司 一种信号识别方法及其设备、存储介质、终端
CN109785850A (zh) * 2019-01-18 2019-05-21 腾讯音乐娱乐科技（深圳）有限公司 一种噪声检测方法、装置和存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106202952A (zh) * 2016-07-19 2016-12-07 南京邮电大学 一种基于机器学习的帕金森疾病诊断方法
CN108073856A (zh) * 2016-11-14 2018-05-25 华为技术有限公司 噪音信号的识别方法及装置
US20190297412A1 (en) * 2018-03-26 2019-09-26 Sony Corporation Apparatuses and Methods for Acoustic Noise Cancelling
CN109903752A (zh) * 2018-05-28 2019-06-18 华为技术有限公司 对齐语音的方法和装置

Also Published As

Publication number Publication date
CN111797708A (zh) 2020-10-20

Similar Documents

Publication Publication Date Title
WO2021248522A1 (zh) 气流杂音检测方法、装置、终端及存储介质
Harb et al. Gender identification using a general audio classifier
US8036884B2 (en) Identification of the presence of speech in digital audio data
CN108198547B (zh) 语音端点检测方法、装置、计算机设备和存储介质
WO2021128741A1 (zh) 语音情绪波动分析方法、装置、计算机设备及存储介质
US11386916B2 (en) Segmentation-based feature extraction for acoustic scene classification
CN107799126A (zh) 基于有监督机器学习的语音端点检测方法及装置
CN111816218A (zh) 语音端点检测方法、装置、设备及存储介质
WO2018095167A1 (zh) 声纹识别方法和声纹识别系统
Jaafar et al. Automatic syllables segmentation for frog identification system
CN109584888A (zh) 基于机器学习的鸣笛识别方法
CN110189767B (zh) 一种基于双声道音频的录制移动设备检测方法
Jena et al. Gender recognition of speech signal using knn and svm
Mohammed et al. Mitigate the reverberant effects on speaker recognition via multi-training
Alonso-Martin et al. Multidomain voice activity detection during human-robot interaction
Bäckström et al. Voice activity detection
Kaminski et al. Automatic speaker recognition using a unique personal feature vector and Gaussian Mixture Models
CN114302301B (zh) 频响校正方法及相关产品
CN112216285B (zh) 多人会话检测方法、系统、移动终端及存储介质
Krijnders et al. Tone-fit and MFCC scene classification compared to human recognition
CN111091816B (zh) 一种基于语音评测的数据处理系统及方法
CN114333874A (zh) 处理音频信号的方法
Hashemi et al. Persian music source separation in audio-visual data using deep learning
WO2021248523A1 (zh) 气流杂音消除方法、装置、计算机设备及存储介质
Agarwal et al. Quantitative analysis of feature extraction techniques for isolated word recognition

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20939493

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20939493

Country of ref document: EP

Kind code of ref document: A1