Audio data enhancement method and system suitable for complex environments

Abstract

The present application relates to the technical field of audio data processing, and specifically to an audio data enhancement method and system suitable for complex environments, the method comprising: obtaining an audio data sequence and a spectrum sequence of audio data in each time frame; obtaining a speech-intensive time frame based on the discrete degree of the amplitude of the audio data sequence; calculating the speech estimation strength of each time frame; dividing the spectrum sequence of each time frame into a preset number of spectrum subsequences at equal intervals; calculating the frequency band interference evaluation value of each spectrum subsequence of the spectrum sequence of each time frame; obtaining the residual noise energy threshold of each spectrum subsequence in the spectrum sequence of each time frame, and enhancing the audio data. The present application improves the enhancement effect of audio data in complex environments.

Description

Technical Field

[0001] The present application relates to the technical field of audio data processing, and in particular to an audio data enhancement method and system suitable for use in complex environments. Background Art

[0002] Audio data enhancement technology refers to the technology of extracting useful signals from background noise and suppressing and reducing noise interference after the audio signal is interfered by various types of noise. As an important part of audio signals, speech signals carry a lot of useful information, but they will inevitably be interfered by the background noise of the surrounding environment, affecting the quality and intelligibility of speech signals. Speech enhancement can enhance the desired signal and suppress irrelevant signals and interference. It is an effective way to solve the throat pollution of speech signals and is also a key link and step in speech signal processing.

[0003] The subspace audio enhancement algorithm decomposes the noisy speech signal space into a signal plus noise subspace and a noise subspace, suppresses the noise signal in the noise subspace, and performs speech prediction in the signal subspace to achieve the purpose of audio data enhancement, and the residual noise generated after enhancement is less. There are many types of interference sources in outdoor recordings, and different interferences are superimposed on each other. Therefore, the noise signal characteristics in the outdoor recording audio data have strong time-varying properties. At present, the subspace audio enhancement algorithm has a good enhancement effect on audio data interfered by fixed-power Gaussian white noise, but when processing audio data of outdoor recordings, it cannot adapt to the strong non-stationarity of outdoor noise signals, and the enhancement effect on audio data is not good. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide an audio data enhancement method and system suitable for complex environments. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides an audio data enhancement method suitable for use in a complex environment, the method comprising the following steps:

[0006] Obtaining audio data sequence and spectrum sequence of audio data in each time frame;

[0007] Acquire a speech intensive time frame based on the discrete degree of the amplitude of the audio data sequence;

[0008] Obtaining the speech estimation intensity of each time frame based on the number of speech-intensive time frames and the amplitude of the audio data sequence;

[0009] Dividing the spectrum sequence of each time frame into a preset number of spectrum subsequences at equal intervals;

[0010] Obtaining a frequency band interference evaluation value of each spectrum subsequence of the spectrum sequence of each time frame based on the speech estimation strength and the amplitude of the spectrum sequence;

[0011] Based on the frequency band interference evaluation value and the amplitude of the spectrum sequence, a residual noise energy threshold of each spectrum subsequence in the spectrum sequence of each time frame is obtained;

[0012] The audio data is enhanced based on a residual noise energy threshold.

[0013] Further, the method for acquiring the speech intensive time frame is:

[0014] Using a threshold segmentation algorithm to process the absolute values ​​of the amplitudes of the audio data of the audio data sequences of all time frames, and obtaining an optimal segmentation threshold as an audio amplitude threshold;

[0015] For the audio data sequence of each time frame, the mean square error of the amplitude of all audio data in the audio data sequence is calculated, and the number of time frames whose mean square error is greater than the audio amplitude threshold is counted as speech-intensive time frames.

[0016] Further, the method for obtaining the speech estimation strength is: ; In the formula, represents the speech estimation intensity of the audio data sequence of the tth time frame; QA represents the number of speech-intensive time frames, represents the absolute value of the amplitude of the nth audio data in the audio data sequence of the tth time frame, N is the length of the audio data sequence of each time frame, AT represents the audio amplitude threshold, It is the preset parameter factor.

[0017] Further, the method for obtaining the frequency band interference evaluation value is:

[0018] Obtaining the spectrum power and spectrum maximum value of each spectrum subsequence based on the amplitude of the spectrum sequence;

[0019] The calculation formula of the frequency band interference evaluation value is: ; In the formula, represents the frequency band interference evaluation value of the mth spectrum subsequence in the spectrum sequence of the tth time frame; Represents the speech estimation intensity of the tth time frame of the audio data sequence; Represents the maximum amplitude value in the spectrum sequence of the tth time frame; Indicates the maximum spectrum value of the mth spectrum subsequence in the tth time frame; Represents the maximum value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; Represents the mean value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; Represents an exponential function with a natural constant as base.

[0020] Furthermore, the method for obtaining the spectrum power is:

[0021] The sum of the squares of the amplitudes in each spectrum subsequence in the spectrum sequence is calculated as the spectrum power of each spectrum subsequence.

[0022] Furthermore, the spectrum maximum is the maximum amplitude value in each spectrum subsequence of the spectrum sequence of each time frame.

[0023] Furthermore, the method for obtaining the residual noise energy threshold is:

[0024] Counting the number of the first spectrum subsequences based on the average of the spectrum power;

[0025] The calculation formula of the residual noise energy threshold is: ; In the formula, Residual noise energy threshold of the mth spectrum subsequence of the spectrum sequence of the tth time frame; represents the frequency band interference evaluation value of the mth spectrum subsequence of the spectrum sequence of the tth time frame; represents the normalization function; Represents the minimum value of the spectrum power of all spectrum data sequences in the spectrum sequence of the tth time frame; represents the number of the first spectrum subsequence; It is the preset parameter factor.

[0026] Furthermore, the method for obtaining the number of the first spectrum subsequences is:

[0027] The mean of the spectrum power of the mth spectrum subsequence of the spectrum sequence of all time frames is calculated as the first mean, and the number of spectrum subsequences whose spectrum power is greater than the first mean in the mth spectrum subsequence of the spectrum sequence of all time frames is counted as the first number of spectrum subsequences.

[0028] Further, the enhancing the audio data based on the residual noise energy threshold comprises:

[0029] The audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is used as the input of the subspace speech enhancement algorithm, and the enhanced audio data of the audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is obtained in combination with the residual noise energy threshold. The enhanced audio data of the audio data corresponding to all spectral subsequences of all spectral sequences of each time frame are arranged in chronological order as the enhanced audio data of each time frame.

[0030] In a second aspect, an embodiment of the present application further provides an audio data enhancement system suitable for complex environments, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0031] This application has at least the following beneficial effects:

[0032] According to the characteristics of the audio data sequence and its different time frames in the time domain and frequency domain, the speech estimation strength and the frequency band interference evaluation value are calculated in turn. The beneficial effect is that by analyzing the difference between the outdoor noise signal and the speech signal in the audio data in the time domain and frequency domain, the accuracy of the calculation of the interference degree of the audio data in different frequency bands is improved, thereby improving the adaptability of the subsequent subspace speech enhancement algorithm to the time-varying outdoor noise. According to the frequency band interference evaluation value, the residual noise energy threshold is calculated. The subspace speech enhancement algorithm is used to obtain enhanced audio data and complete the enhancement of audio data in complex environments. The beneficial effect is that the difference in the non-stationarity of outdoor noise signals and speech signals on different time scales is used to adjust the residual noise energy threshold in the subspace speech enhancement algorithm in real time, and perform data enhancement on different frequency bands of the audio signal respectively. Improve the adaptability to the time-varying outdoor noise, reduce the erroneous suppression of speech signal details, avoid distortion of speech signals, and improve the enhancement effect of audio data in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 A flowchart of a method for enhancing audio data in a complex environment provided by an embodiment of the present application;

[0035] Figure 2 A flowchart for obtaining a residual noise energy threshold is provided for one embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, describes in detail the specific implementation methods, structures, features and effects of the audio data enhancement method and system for complex environments proposed in the present application. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0038] The following is a detailed description of the specific scheme of the audio data enhancement method and system applicable to complex environments provided by the present application with reference to the accompanying drawings.

[0039] See also Figure 1 , which shows a flowchart of a method for enhancing audio data in a complex environment provided by an embodiment of the present application, the method comprising the following steps:

[0040] Step S1, obtaining an audio data sequence and a spectrum sequence of audio data in each time frame.

[0041] For each audio data to be enhanced, the audio data of the audio file is read in sequence, and the audio data per second is organized into an audio data sequence in chronological order. The sampling frequency of the audio file in this application is 48000 Hz, and the length of each audio data sequence is 48000.

[0042] The speech signal is a non-stationary random process, and the characteristics of the speech signal have strong time-varying properties. However, in a shorter time range, the characteristics of the speech signal are basically stable, and the speech signal can be regarded as a quasi-stationary process. Therefore, the audio data sequence is framed, and the audio data sequence is equally spaced into T time frames to obtain the audio data sequence of each time frame. The length of each time frame is N. In this embodiment, T and N are 50 and 960 respectively. The implementer can select other values ​​according to the actual situation.

[0043] In order to obtain the frequency distribution of audio data in the time frame, the audio data of each time frame is processed using fast Fourier transform to obtain the spectrum sequence of the audio data of each time frame, where the length of the spectrum sequence of each time frame is N. In this embodiment, the value of N is 960, and the implementer can select other values ​​according to actual conditions.

[0044] Step S2, obtaining speech-intensive time frames based on the discrete degree of the amplitude of the audio data sequence; obtaining the speech estimation strength of each time frame based on the number of speech-intensive time frames and the amplitude of the audio data sequence; dividing the spectrum sequence of each time frame into a preset number of spectrum sub-sequences at equal intervals; obtaining the frequency band interference evaluation value of each spectrum sub-sequence of the spectrum sequence of each time frame based on the speech estimation strength and the amplitude of the spectrum sequence.

[0045] When recording outdoors, the distribution of speech signals in the time domain is discontinuous, while outdoor wind noise and other noise sources continue to exist, and their intensity distribution is continuous in the time domain. At the same time, the distance between the shooting or recording personnel and the recording equipment is relatively close. Therefore, compared with outdoor noise, the speech signal intensity in the audio data is greater.

[0046] Further, the absolute value of the amplitude of the audio data of the audio data sequence of all time frames is processed using a threshold segmentation algorithm to obtain an optimal segmentation threshold as the audio amplitude threshold. The threshold segmentation algorithm selected in this embodiment is the Otsu threshold segmentation algorithm, and the implementer can select other threshold segmentation algorithms according to actual conditions. Among them, the threshold segmentation algorithm is a well-known technology and will not be described in detail in this embodiment.

[0047] Because the strength of speech signals during outdoor recording is significantly greater than the strength of outdoor noise signals, and the threshold segmentation algorithm can divide the absolute value of audio data into two types of data with obvious differences in size distribution through threshold segmentation, the two types of data obtained by classification represent the audio data of noise and speech respectively, so as to further estimate the speech signal strength in different time frames in the current collected audio data sequence.

[0048] Furthermore, for the audio data sequence of each time frame, the mean square error of the amplitude of all audio data in the audio data sequence is calculated, and the number of time frames in which the mean square error is greater than the audio amplitude threshold is counted as speech-intensive time frames.

[0049] According to the above analysis, in order to reflect the richness of speech information contained in the time frame, based on the number of speech-intensive time frames and the amplitude of the audio data sequence, the speech estimation strength of each time frame is obtained. The calculation formula is: ; In the formula, represents the speech estimation intensity of the audio data sequence of the tth time frame; QA represents the number of speech-intensive time frames, represents the absolute value of the amplitude of the nth audio data in the audio data sequence of the tth time frame, N is the length of the audio data sequence of each time frame, AT represents the audio amplitude threshold, It is a preset parameter adjustment factor, and its value in this embodiment is 1, in order to prevent the number of speech-intensive time frames from being 0, resulting in the value of the speech estimation intensity being 0.

[0050] It should be noted that the root mean square value of the audio data in the time frame reflects the average signal strength of the outdoor collected audio signal. The more time frames in the audio data sequence whose average signal strength exceeds the audio amplitude threshold, that is, the more speech-intensive time frames there are, the denser the distribution of speech signals in the audio data sequence, the richer the overall speech information contained in the sequence, and the greater the speech estimation strength. In addition, the larger the ratio of the absolute value of the audio data in the time frame to the audio amplitude threshold, the more speech components are contained in the time frame and the greater the strength of the speech signal, and the greater the speech estimation strength obtained by calculation.

[0051] Since there are multiple types of interference sources in outdoor recording, the noise frequency distribution and interference duration of different interference sources vary greatly, making the outdoor noise signal highly non-stationary. At the same time, wind noise is the main noise source for outdoor recording, and its signal characteristics have strong time-varying properties, which aggravates the randomness of the interference to the voice signal.

[0052] Since outdoor noise signals have continuity in frequency distribution in a short time, their non-stationarity is weaker than that of speech signals, and they can be treated as stationary signals in audio data sequences. Speech signals can be treated as stationary signals in a shorter time frame, but in audio data sequences, their signal characteristics are more time-varying, and there is a big difference in the changes in frequency distribution from noise signals.

[0053] For the spectrum sequence of each time frame, the spectrum sequence is divided into M spectrum subsequences at equal intervals. In this embodiment, the value of M is 16, and the implementer can select other values ​​according to actual conditions. The sum of the squares of the amplitudes in each spectrum subsequence in the spectrum sequence is calculated as the spectrum power of each spectrum subsequence, which represents the power of the frequency component of the time frame in the corresponding frequency interval; further, the maximum value of the amplitude in each spectrum subsequence of the spectrum sequence of each time frame is taken as the spectrum maximum value of each spectrum subsequence.

[0054] Furthermore, in order to reflect the degree of interference of the audio data by the outdoor environmental noise, the frequency band interference evaluation value of each spectrum subsequence of the spectrum sequence of each time frame is obtained based on the speech estimation strength and the amplitude of the spectrum sequence. The calculation formula is: ; In the formula, represents the frequency band interference evaluation value of the mth spectrum subsequence in the spectrum sequence of the tth time frame, reflecting the interference degree of the spectrum subsequence caused by the outdoor environmental noise; Represents the speech estimation intensity of the tth time frame of the audio data sequence; Represents the maximum amplitude value in the spectrum sequence of the tth time frame; Indicates the maximum spectrum value of the mth spectrum subsequence in the tth time frame; Represents the maximum value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; Represents the mean value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; It represents an exponential function with the natural constant e as the base, and its purpose is to normalize the difference between the maximum value of the spectrum power and the average value of the spectrum power, and at the same time represents its negative correlation with the frequency band interference assessment value.

[0055] It should be noted that the speech estimation strength reflects the signal strength of the speech signal in the time frame. The smaller the speech estimation strength, the greater the relative interference degree of the noise of the same power to the speech signal, and the greater the frequency band interference evaluation value. Since the power of the outdoor noise signal is small, in the time frame of the speech signal distribution, the maximum spectrum amplitude in the spectrum sequence reflects the main frequency amplitude of the speech signal in the time frame. The greater the difference between the spectrum peak of the spectrum subsequence and the maximum spectrum amplitude of the spectrum sequence, the less speech information is contained in the corresponding frequency component of the spectrum subsequence, the greater the relative interference degree of the noise in this frequency band, and the greater the frequency band interference evaluation value. In addition, for the spectrum subsequence with more speech signals and larger speech signal power, due to the strong non-stationarity of the speech signal in the audio data sequence, its spectrum power distribution changes greatly over time. The average spectrum power of the spectrum subsequence corresponding to N time frames in the audio data sequence is greatly different from the maximum spectrum power. The probability of this frequency band containing speech information is greater and the speech signal power is greater. The interference degree of outdoor environmental noise is smaller, and the calculated frequency band interference evaluation value is smaller.

[0056] Step S3, based on the frequency band interference evaluation value and the amplitude of the spectrum sequence, obtaining the residual noise energy threshold of each spectrum subsequence in the spectrum sequence of each time frame; and enhancing the audio data based on the residual noise energy threshold.

[0057] The subspace speech enhancement algorithm decomposes the vector space of the noisy speech signal into two mutually orthogonal subspaces, one of which contains only noise and the other contains both speech and noise. By setting the noise subspace to zero, an approximate estimate of the pure speech signal is obtained. When constructing the constraint estimator, the subspace speech enhancement algorithm controls the residual noise energy of different components below the residual noise energy threshold to ensure that the speech signal distortion is minimized. Due to the different non-stationary characteristics of outdoor noise and speech signals, the degree of aliasing of noise and speech in different frequency bands is different. Therefore, it is necessary to determine the residual noise energy threshold for subspace speech enhancement of audio signals in different frequency bands.

[0058] The mean of the spectrum power of the mth spectrum subsequence of the spectrum sequence of all time frames is calculated as the first mean, and the number of spectrum subsequences whose spectrum power is greater than the first mean in the mth spectrum subsequence of the spectrum sequence of all time frames is counted as the first number of spectrum subsequences.

[0059] Therefore, according to the above analysis, based on the average situation of spectrum power, the frequency band interference assessment value and the amplitude of the spectrum sequence, the residual noise energy threshold of each spectrum subsequence in the spectrum sequence of each time frame is obtained, and the calculation formula is: ; In the formula, The residual noise energy threshold of the mth spectrum subsequence of the spectrum sequence of the tth time frame reflects the degree of noise suppression by the subspace speech enhancement algorithm; represents the frequency band interference evaluation value of the mth spectrum subsequence of the spectrum sequence of the tth time frame; represents the normalization function; Represents the minimum value of the spectrum power of all spectrum data sequences in the spectrum sequence of the tth time frame; Indicates the number of the first spectrum subsequence, reflecting the probability of the speech signal being distributed in the spectrum subsequence; is a preset parameter adjustment factor, which is set to 1 in this embodiment to prevent the denominator from being zero. Figure 2 shown.

[0060] It should be noted that the minimum value of the spectrum energy of the spectrum subsequence within the time frame is used as the minimum noise power within the time frame. On the one hand, the smaller the frequency band interference evaluation value is, the more it indicates that the power of the speech signal is greater than the noise power. At this time, the threshold can be set to be smaller to further suppress the noise. The larger the frequency band interference evaluation value is, the higher the noise power is. At this time, the threshold needs to be set to be larger to avoid excessive filtering of the speech signal in this frequency band, retain the details of the speech signal, and improve the quality of speech enhancement. On the other hand, in the audio data sequence, the more time frames the spectrum power of the frequency domain subsequence exceeds the average spectrum power of the spectrum sequence, the higher the probability that the speech signal is distributed in the corresponding frequency band. A smaller threshold is set to improve the degree of noise suppression for the spectrum subsequence containing more speech frequency components.

[0061] At this point, the residual noise energy threshold of each spectral subsequence of the spectral sequence of each time frame is obtained, and the audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is used as the input of the subspace speech enhancement algorithm. The enhanced audio data of the audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is obtained in combination with the residual noise energy threshold. The enhanced audio data of the audio data corresponding to all spectral subsequences of all spectral sequences of all time frames are arranged in chronological order as the enhanced audio data of each time frame, thereby completing the enhancement of the audio data.

[0062] Based on the same inventive concept as the above method, an embodiment of the present application also provides an audio data enhancement system suitable for use in complex environments, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of any one of the above-mentioned audio data enhancement methods suitable for use in complex environments are implemented.

[0063] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0065] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application should be included in the protection scope of the present application.

Claims

1. An audio data enhancement method suitable for complex environments, characterized in that: The method comprises the following steps: Obtaining audio data sequence and spectrum sequence of audio data in each time frame; Acquire a speech intensive time frame based on the discrete degree of the amplitude of the audio data sequence; Obtaining the speech estimation intensity of each time frame based on the number of speech-intensive time frames and the amplitude of the audio data sequence; Dividing the spectrum sequence of each time frame into a preset number of spectrum subsequences at equal intervals; Obtaining a frequency band interference evaluation value of each spectrum subsequence of the spectrum sequence of each time frame based on the speech estimation strength and the amplitude of the spectrum sequence; Based on the frequency band interference evaluation value and the amplitude of the spectrum sequence, a residual noise energy threshold of each spectrum subsequence in the spectrum sequence of each time frame is obtained; The audio data is enhanced based on a residual noise energy threshold.

2. The audio data enhancement method suitable for complex environments according to claim 1, characterized in that: The method for acquiring the speech intensive time frame is: Using a threshold segmentation algorithm to process the absolute values ​​of the amplitudes of the audio data of the audio data sequences of all time frames, and obtaining an optimal segmentation threshold as an audio amplitude threshold; For the audio data sequence of each time frame, the mean square error of the amplitude of all audio data in the audio data sequence is calculated, and the number of time frames whose mean square error is greater than the audio amplitude threshold is counted as speech-intensive time frames.

3. The audio data enhancement method suitable for complex environments according to claim 1, characterized in that: The method for obtaining the speech estimation strength is: ; In the formula, represents the speech estimation intensity of the audio data sequence of the tth time frame; QA represents the number of speech-intensive time frames, represents the absolute value of the amplitude of the nth audio data in the audio data sequence of the tth time frame, N is the length of the audio data sequence of each time frame, AT represents the audio amplitude threshold, It is the preset parameter factor.

4. The audio data enhancement method suitable for complex environments according to claim 1, characterized in that: The method for obtaining the frequency band interference evaluation value is: Obtaining the spectrum power and spectrum maximum value of each spectrum subsequence based on the amplitude of the spectrum sequence; The calculation formula of the frequency band interference evaluation value is: ; In the formula, represents the frequency band interference evaluation value of the mth spectrum subsequence in the spectrum sequence of the tth time frame; Represents the estimated speech intensity of the tth time frame of the audio data sequence; Represents the maximum amplitude value in the spectrum sequence of the tth time frame; Indicates the maximum spectrum value of the mth spectrum subsequence in the tth time frame; Represents the maximum value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; Represents the mean value of the spectrum power of the mth spectrum subsequence in the spectrum sequence of all time frames; Represents an exponential function with a natural constant as base.

5. The audio data enhancement method suitable for complex environments as claimed in claim 4, characterized in that: The method for obtaining the spectrum power is: The sum of the squares of the amplitudes in each spectrum subsequence in the spectrum sequence is calculated as the spectrum power of each spectrum subsequence.

6. The audio data enhancement method suitable for complex environments as claimed in claim 4, characterized in that: The spectrum maximum is the maximum amplitude value in each spectrum subsequence of the spectrum sequence of each time frame.

7. The audio data enhancement method suitable for complex environments as claimed in claim 4, characterized in that: The method for obtaining the residual noise energy threshold is: Counting the number of the first spectrum subsequences based on the average of the spectrum power; The calculation formula of the residual noise energy threshold is: ; In the formula, Residual noise energy threshold of the mth spectrum subsequence of the spectrum sequence of the tth time frame; represents the frequency band interference evaluation value of the mth spectrum subsequence of the spectrum sequence of the tth time frame; represents the normalization function; Represents the minimum value of the spectrum power of all spectrum data sequences in the spectrum sequence of the tth time frame; represents the number of the first spectrum subsequence; It is the preset parameter factor.

8. The audio data enhancement method suitable for complex environments according to claim 7, characterized in that: The method for obtaining the number of the first spectrum subsequences is: The mean of the spectrum power of the mth spectrum subsequence of the spectrum sequence of all time frames is calculated as the first mean, and the number of spectrum subsequences whose spectrum power is greater than the first mean in the mth spectrum subsequence of the spectrum sequence of all time frames is counted as the first number of spectrum subsequences.

9. The audio data enhancement method suitable for complex environments according to claim 1, characterized in that: The step of enhancing the audio data based on the residual noise energy threshold comprises: The audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is used as the input of the subspace speech enhancement algorithm, and the enhanced audio data of the audio data corresponding to each spectral subsequence of the spectral sequence of each time frame is obtained in combination with the residual noise energy threshold. The enhanced audio data of the audio data corresponding to all spectral subsequences of all spectral sequences of each time frame are arranged in chronological order as the enhanced audio data of each time frame.

10. An audio data enhancement system suitable for use in complex environments, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the audio data enhancement method suitable for complex environments are implemented as described in any one of claims 1 to 9.

Images