Audio signal anomaly detection method based on isolated forest algorithm and processing terminal

Abstract

The invention discloses an audio signal anomaly detection method based on an isolated forest algorithm. The audio signal anomaly detection method comprises the following steps: step 1, acquiring an audio signal; 2, performing feature extraction on the audio signal, generating a feature vector, and obtaining an audio feature vector; 3, inputting the audio feature vectors into an isolated forest model, and training an isolated forest by taking the audio feature vectors as training data to obtain a trained isolated forest model; and 4, receiving a to-be-detected target audio signal, performing feature extraction on the target audio signal to obtain a target audio feature vector, and calculating an abnormal score of the target audio feature vector through the isolated forest model to judge whether the target audio signal is abnormal or not. According to the unsupervised learning algorithm based on the isolated forest algorithm, rules do not need to be set manually, only one preset threshold value for abnormal score comparison needs to be set, and other threshold values do not need to be set.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal anomaly detection, and in particular to an audio signal anomaly detection method and a processing terminal based on an isolation forest algorithm. Background Art

[0002] There are three main technical paths to detect abnormal audio signals in the existing technology:

[0003] 1. Threshold detection: This type of method determines whether the audio signal is abnormal by artificially setting a threshold. When the amplitude of the audio signal exceeds the threshold, it is considered abnormal.

[0004] Second, detection based on statistical models: This type of method collects a large number of audio signals and builds models to statistically analyze the characteristics of the audio signals through modeling, and then determines whether the audio signals are abnormal based on the characteristics.

[0005] 3. Detection based on deep learning: This type of method uses deep learning to learn features from a large amount of data, and then judges whether the audio signal is abnormal based on the learned features.

[0006] The existing detection methods under these three technical paths all have a common shortcoming, that is, they cannot adapt to abnormal detection under dynamically changing audio signals. In addition, most of the existing methods rely heavily on manually set rules or models, which makes it difficult to deal with complex audio signal abnormal scenarios. Summary of the invention

[0007] In view of the deficiencies in the prior art, an object of the present invention is to provide an audio signal anomaly detection method and a processing terminal based on an isolation forest algorithm, which can solve the problems described in the background technology.

[0008] The technical solution to achieve the purpose of the present invention is: an audio signal anomaly detection method based on an isolation forest algorithm, comprising the following steps:

[0009] Step 1: Obtain audio signal through acquisition;

[0010] Step 2: extracting features from the audio signal to obtain audio features, and generating feature vectors from the extracted audio features to obtain audio feature vectors;

[0011] Step 3: Input the audio feature vector into the isolation forest model to train the isolation forest using the audio feature vector as training data. During the training process, each audio feature vector is used as a data point. Multiple decision trees are set in the isolation forest model. The decision trees process the data points in isolation and calculate the abnormality score of each audio feature vector based on the path length of each data point. Thus, the training is completed and the trained isolation forest model is obtained.

[0012] Step 4: Receive the target audio signal to be detected, extract the features of the target audio signal, obtain the target audio features of the target audio signal, and generate a feature vector from the target audio features to obtain the target audio feature vector.

[0013] The target audio feature vector is input into the trained isolation forest model, and the abnormality score of the target audio feature vector is calculated by the isolation forest model. If the score result exceeds the preset threshold, the target audio signal corresponding to the target audio feature vector is judged to be abnormal.

[0014] Furthermore, in step 2, the collected audio signal is a time domain audio signal, and the time domain audio signal is converted into a frequency domain audio signal to obtain the MFCC and spectrum of the frequency domain audio signal, and the mean and variance of the audio signal in the time domain are calculated, and a feature vector is generated based on the audio signals in the time domain and frequency domain, and finally an audio feature vector is obtained.

[0015] Furthermore, after step 1 and before step 2, the method further includes preprocessing the audio signal to obtain a preprocessed audio signal, and step 2 extracts features from the preprocessed audio signal.

[0016] Furthermore, the preprocessing includes filtering and / or denoising, and the denoising includes removing background noise.

[0017] Furthermore, after step 4, the method further includes issuing an alarm and handling the alarm, issuing an alarm signal when it is determined that an abnormal target audio signal exists,

[0018] When an alarm signal is received, an alarm process is performed, and the alarm process includes stopping the output of audio and / or triggering a restart operation.

[0019] Furthermore, the method also includes recording and storing the detection results of the target audio signal and generating a visual report.

[0020] A processing terminal, comprising:

[0021] A memory for storing program instructions;

[0022] The processor is used to run the program instructions to execute the steps of the audio signal anomaly detection method based on the isolation forest algorithm.

[0023] Beneficial effects of the present invention: The present invention is based on an unsupervised learning algorithm of the isolation forest algorithm, which does not require manual setting of rules. It only requires setting a preset threshold for abnormal score comparison, and does not require setting other thresholds. The isolation forest algorithm can automatically identify and process abnormal audio signals, reduce manual intervention, and has strong real-time performance, and can quickly respond to abnormal audio signals and trigger alarms and alarm disposal.

[0024] Among them, the audio feature vector obtained by extracting the audio signal features is input into the isolation forest algorithm and trained based on the features, so that the trained isolation forest algorithm can adapt to complex and changeable audio signal data, and can avoid the problem of abnormal misjudgment or missed judgment caused by inaccurate human settings in traditional processing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of a flow chart of a preferred embodiment;

[0026] Figure 2 A schematic diagram of the structure of the processing terminal. DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:

[0028] like Figure 1 As shown, a method for detecting anomalies in an audio signal based on an isolation forest algorithm comprises the following steps:

[0029] Step 1: Obtain audio signal through acquisition.

[0030] The audio signal may be acquired in real time through a microphone or an audio sensor, or through other means. The audio signal may be acquired temporarily based on the present method, or through previously acquired audio signals.

[0031] Step 2: extracting features from the audio signal to obtain audio features, and generating feature vectors from the extracted audio features to obtain audio feature vectors.

[0032] The collected audio signal is a time domain audio signal, which is converted into a frequency domain audio signal. For example, the MFCC and spectrum of the frequency domain audio signal are obtained through time domain-frequency domain conversion processing. The mean and variance of the audio signal can be calculated in the time domain, and a feature vector is generated based on the audio signals in the time domain and frequency domain, and finally an audio feature vector is obtained.

[0033] Exemplarily, after step 1 and before step 2, the audio signal is preprocessed to obtain a preprocessed audio signal, and step 2 extracts features from the preprocessed audio signal. Preprocessing includes filtering and / or denoising, and denoising mainly removes background noise to improve the quality of the audio signal.

[0034] Wherein, preprocessing may be performed by a high-pass filter and / or a low-pass filter to remove noise.

[0035] Step 3: Input the audio feature vector into the isolation forest model to train the isolation forest using the audio feature vector as training data. During the training process, each audio feature vector is used as a data point. Multiple decision trees are set in the isolation forest model. The decision trees process the data points in isolation and calculate the anomaly score of each audio feature vector based on the path length of each data point, thereby completing the training and obtaining the trained isolation forest model.

[0036] The Isolation Forest Algorithm calculates the score of each data point based on the path length and evaluates the degree of isolation of each data point by constructing a decision tree. The path length h(x) of a data point x represents the depth from the root node to the data point x in the decision tree. Outliers are often isolated earlier, so they have a shorter path length.

[0037] The calculation formula for the anomaly score is as follows:

[0038]

[0039] Where s(x) represents the anomaly score of data x, E(h(x)) represents the average path length of data point x in the isolation forest model, c(n) represents the expected path length of the data set consisting of normal data points, and n represents the amount of data in the data set consisting of normal data points.

[0040] The larger s(x) is, the higher the anomaly score is, and the more likely the data point is an anomaly.

[0041] Step 4: Receive a target audio signal to be detected, perform feature extraction on the target audio signal to obtain a target audio feature of the target audio signal, and generate a feature vector from the target audio feature to obtain a target audio feature vector.

[0042] The target audio feature vector is input into the trained isolation forest model, and the abnormality score of the target audio feature vector is calculated by the isolation forest model. If the score result exceeds the preset threshold, the target audio signal corresponding to the target audio feature vector is judged to be abnormal.

[0043] Exemplarily, after step 4, it also includes issuing an alarm and alarm handling, and issuing an alarm signal after determining that there is an abnormal target audio signal. After receiving the alarm signal, an alarm handling is performed, and the alarm handling includes stopping audio output, triggering restart, and other operations.

[0044] Exemplarily, the method also includes recording and storing the detection results of the target audio signal, and generating a visual report to display the detection history, abnormal frequency, type and other information.

[0045] The present invention is based on an unsupervised learning algorithm of the isolation forest algorithm. It does not require manual setting of rules. It only requires setting a preset threshold for abnormal score comparison, and does not require setting other thresholds. The isolation forest algorithm can automatically identify and process abnormal audio signals, reduce manual intervention, and has strong real-time performance. It can quickly respond to abnormal audio signals and trigger alarms and alarm disposal.

[0046] Among them, the audio feature vector obtained by extracting the audio signal features is input into the isolation forest algorithm and trained based on the features, so that the trained isolation forest algorithm can adapt to complex and changeable audio signal data, and can avoid the problem of abnormal misjudgment or missed judgment caused by inaccurate human settings in traditional processing methods.

[0047] like Figure 2 As shown, the present invention also provides a processing terminal 100, which includes:

[0048] Memory 101, used for storing program instructions;

[0049] The processor 102 is used to run the program instructions to execute the steps of the audio signal anomaly detection method based on the isolation forest algorithm.

[0050] The embodiment disclosed in this specification is only an example of a unilateral feature of the present invention, and the protection scope of the present invention is not limited to this embodiment, and any other functionally equivalent embodiments fall within the protection scope of the present invention. For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A method for detecting anomalies in audio signals based on an isolation forest algorithm, characterized in that: The following steps are involved: Step 1: Obtain audio signal through acquisition; Step 2: extracting features from the audio signal to obtain audio features, and generating feature vectors from the extracted audio features to obtain audio feature vectors; Step 3: Input the audio feature vector into the isolation forest model to train the isolation forest using the audio feature vector as training data. During the training process, each audio feature vector is used as a data point. Multiple decision trees are set in the isolation forest model. The decision trees process the data points in isolation and calculate the abnormality score of each audio feature vector based on the path length of each data point. Thus, the training is completed and the trained isolation forest model is obtained. Step 4: Receive the target audio signal to be detected, extract the features of the target audio signal, obtain the target audio features of the target audio signal, and generate a feature vector from the target audio features to obtain the target audio feature vector. The target audio feature vector is input into the trained isolation forest model, and the abnormality score of the target audio feature vector is calculated by the isolation forest model. If the score result exceeds the preset threshold, the target audio signal corresponding to the target audio feature vector is judged to be abnormal.

2. The audio signal anomaly detection method based on the isolation forest algorithm according to claim 1 is characterized in that: In step 2, the collected audio signal is a time domain audio signal. The time domain audio signal is converted into a frequency domain audio signal to obtain the MFCC and spectrum of the frequency domain audio signal, and the mean and variance of the audio signal in the time domain are calculated. A feature vector is generated based on the audio signals in the time domain and frequency domain, and finally an audio feature vector is obtained.

3. The audio signal anomaly detection method based on the isolation forest algorithm according to claim 2 is characterized in that: After step 1 and before step 2, the method further includes preprocessing the audio signal to obtain a preprocessed audio signal. Step 2 extracts features from the preprocessed audio signal.

4. The audio signal anomaly detection method based on the isolation forest algorithm according to claim 1 or 3, characterized in that: Preprocessing includes filtering and / or denoising, and denoising includes removing background noise.

5. The audio signal anomaly detection method based on the isolation forest algorithm according to claim 1 is characterized in that: After step 4, the method further includes issuing an alarm and handling the alarm. When it is determined that there is an abnormal target audio signal, an alarm signal is issued. When an alarm signal is received, an alarm process is performed, and the alarm process includes stopping the output of audio and / or triggering a restart operation.

6. The audio signal anomaly detection method based on the isolation forest algorithm according to claim 5 is characterized in that: It also includes recording and storing the detection results of the target audio signal and generating a visual report.

7. A processing terminal, characterized in that: It includes: A memory for storing program instructions; A processor is used to run the program instructions to execute the steps of the audio signal anomaly detection method based on the isolation forest algorithm as described in any one of claims 1 to 6.

Images