CN115798502B - Audio denoising method for Bluetooth headset - Google Patents
Audio denoising method for Bluetooth headset Download PDFInfo
- Publication number
- CN115798502B CN115798502B CN202310043255.5A CN202310043255A CN115798502B CN 115798502 B CN115798502 B CN 115798502B CN 202310043255 A CN202310043255 A CN 202310043255A CN 115798502 B CN115798502 B CN 115798502B
- Authority
- CN
- China
- Prior art keywords
- audio
- value
- fitting
- sampling
- fitting deviation
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention relates to the technical field of voice processing, in particular to an audio denoising method for a Bluetooth headset. According to the method, a trend item curve is obtained by carrying out amplitude fitting on an obtained audio signal, audio paragraphs are divided according to the trend item curve, fitting deviation values are obtained according to amplitude differences in the audio paragraphs, first sampling density is obtained by analyzing the difference of the fitting deviation values of the audio paragraphs, second sampling density is obtained by analyzing probability density distribution conditions of different fitting deviation values of the audio paragraphs in the whole audio signal, sampling point number of the audio paragraphs is obtained according to sampling density coefficients obtained by the first sampling density and the second sampling density, and audio is sampled and denoised according to the sampling point number. According to the invention, through analyzing the audio data, when the sampling points are put in each audio paragraph, the self-adaptive increase based on the minimum sampling point number is realized, the sampling point number is reduced to the greatest extent, and the quality and efficiency of audio denoising are improved.
Description
Technical Field
The invention relates to the technical field of voice processing, in particular to an audio denoising method for a Bluetooth headset.
Background
The Bluetooth headset is convenient to carry and use, but has the same disadvantages, no matter how advanced the Bluetooth technology is, the transmission performance of the Bluetooth headset is always weaker than that of wired transmission, the problems of audio data loss, damage, noise and the like often occur in the transmission process, the audio data is distorted, and the decoding chip of the mobile phone also can generate slight noise in the audio analysis process. If the environment of the user is complex, for example, strong electromagnetic interference exists in the outside, even noise of the Bluetooth headset, which causes the Bluetooth headset to produce sound. The influence on the external environment cannot be completely avoided, but the distortion problem of audio data can be reduced by optimizing the transmission mode of the audio, and because the audio signal can accurately determine the signal value at any time in time sequence, the existing wired and wireless transmission of the audio is realized by using a sampling coding mode, so that the method for improving the audio quality generally considers how to scientifically and reasonably adjust the distribution density of sampling points during coding transmission, not only ensures the signal reduction degree, but also can control the transmission time to reduce noise interference.
The quality of the audio signal depends to some extent on the magnitude of the sampling frequency, the higher the signal reduction degree, but correspondingly the transmission time of the audio data is increased, and the longer the transmission time, the more the noise is disturbed. In the existing method for carrying out audio denoising on sampling points, the sampling points are mostly analyzed to reduce the number of the sampling points, so that redundant noise points are removed, but the audio processing method does not reduce the number of the sampling points to the greatest extent, so that the reduced coding efficiency is reduced, the transmission time is prolonged, and the noise attachment rate in transmission is high.
Disclosure of Invention
In order to solve the technical problems that the number of sampling points is not reduced to the greatest extent in the audio processing in the prior art, so that the reduced coding efficiency is reduced, the transmission time is prolonged, and the noise attachment rate is high in the transmission process, the invention aims to provide an audio denoising method for a Bluetooth headset, and the adopted technical scheme is as follows:
the invention provides an audio denoising method for a Bluetooth headset, which comprises the following steps:
acquiring an audio signal, and dividing an audio paragraph according to a trend item curve obtained by fitting the amplitude of the audio signal; obtaining a fitting deviation value of each audio segment according to the audio signal and the trend item curve;
obtaining a deviation value set consisting of fitting deviation values in each audio paragraph according to the distribution of the fitting deviation values, and obtaining a first sampling density of each audio paragraph according to the difference of the fitting deviation values in the deviation value set;
obtaining probability density distribution of each type of fitting deviation values according to the type density distribution conditions of all fitting deviation values in the audio signal; taking any type of fitting deviation values in the deviation value set of any one audio paragraph as reference fitting deviation values, acquiring type influence values of the reference fitting deviation values through the duty ratio of the reference fitting deviation values and the corresponding probability density distribution, and acquiring second sampling density of the audio paragraph according to the type influence values of the fitting deviation values of all types;
obtaining a sampling density coefficient according to the first sampling density and the second sampling density, and obtaining the number of sampling points of each audio paragraph according to the duty ratio of each audio paragraph in the audio signal and the sampling density coefficient; and sampling and denoising the audio according to the number of the sampling points.
Further, the obtaining of the deviation value set includes:
and arranging the fitting deviation values in the audio paragraphs according to a normal distribution mode to obtain a deviation value set and a serial number of each fitting deviation value in the deviation value set.
Further, the obtaining of the first sampling density includes:
obtaining the maximum fitting deviation value and the corresponding maximum sequence number in the deviation value set; in the deviation value set, the minimum fitting deviation value at the left side of the maximum fitting deviation value is obtained and marked as a left minimum fitting deviation value, the serial number corresponding to the left minimum fitting deviation value is marked as a left minimum serial number, the minimum fitting deviation value at the right side of the maximum fitting deviation value is obtained and marked as a right minimum fitting deviation value, and the serial number corresponding to the right minimum fitting deviation value is marked as a right minimum serial number;
taking the average value of the left minimum fitting deviation value and the right minimum fitting deviation value as an average minimum deviation value, and taking the difference value of the maximum fitting deviation value and the average minimum deviation value as a deviation difference value; taking the absolute value of the difference between the maximum sequence number and the left minimum sequence number as a left sequence difference, taking the absolute value of the difference between the maximum sequence number and the right minimum sequence number as a right sequence difference, and taking the average value of the left sequence difference and the right sequence difference as a sequence difference value;
the ratio of the deviation error value to the sequence error value is taken as a first sampling density.
Further, the obtaining of the probability density distribution includes:
counting the types and the quantity of all fitting deviation values in the audio signal to obtain a distribution probability histogram of the fitting deviation values, and obtaining the type distribution probability of each type of fitting deviation values according to the distribution probability histogram; the result obtained by taking each type of distribution probability as an independent variable of the Gaussian probability density function is taken as probability density distribution of the corresponding class fitting deviation value.
Further, the obtaining of the type influence value includes:
taking the result of the fixed integral of probability density distribution corresponding to the reference fitting deviation value as a first reference value;
and taking the ratio of the number of the reference fitting deviation values in the audio paragraph to the total number of the reference fitting deviation values in the audio signal as a number duty ratio, and multiplying the number duty ratio by the first reference value to obtain the type influence value of the reference fitting deviation values.
Further, the obtaining of the second sampling density includes:
normalizing the type influence value of each type of fitting deviation value in the audio paragraph to obtain a corresponding normalized type influence value, and taking the addition result of the normalized type influence values as the second sampling density of the audio paragraph.
Further, the obtaining of the sampling density coefficient includes:
taking the average value of the first sampling density and the second sampling density as a sampling density coefficient.
Further, the obtaining the number of sampling points of each audio paragraph includes:
obtaining the lowest sampling point number of the audio signal, and multiplying the lowest sampling point number by the length ratio of the audio paragraph in the audio signal for any audio paragraph to obtain the lowest sampling point number of the audio paragraph; and obtaining sampling distribution weights according to the sampling density coefficients, multiplying the lowest sampling point number of the paragraphs by the sampling distribution weights, and obtaining the sampling point number of the audio paragraphs.
Further, the obtaining of the audio paragraph includes:
taking the amplitude of the audio signal as a data point, and adopting a least square method to perform curve fitting to obtain a trend item curve; and obtaining extreme values in the trend item curve, dividing a time sequence paragraph between every two adjacent extreme values into an audio paragraph, and obtaining all the audio paragraphs in the audio signal.
Further, the obtaining of the fitting deviation value includes:
and obtaining the signal amplitude of each moment in the audio signal, obtaining the fitting value of each moment on the trend term curve, and taking the difference value of the signal amplitude and the fitting value corresponding to each moment as the fitting deviation value.
The invention has the following beneficial effects:
the invention mainly divides the audio frequency paragraph obtained by the amplitude fitting trend item curve of the audio frequency signal, wherein the fitting deviation value in each audio frequency paragraph can embody the discrete characteristic of the audio frequency signal on the corresponding time sequence of the audio frequency paragraph. While discrete features reflect the distribution of sound features in the audio segment, the more discrete the sound features of the audio segment are, the more densely the number of samples is needed. Therefore, the difference of fitting deviation values in each audio paragraph and the probability density distribution of different fitting deviation values in each audio paragraph in the whole audio signal are analyzed, and the sampling density coefficient of the audio paragraph, namely the retention weight of the sampling point, is obtained through comprehensive calculation, so that the transmission of the retained audio signal is completed more completely. The number of the sampling points is adaptively increased according to the density coefficient of the sampling points of each audio paragraph, so that the number of the sampling points is reduced to the greatest extent, the integrity of the audio is ensured, the sampling efficiency is greatly improved, the transmission time is shortened, and the noise attachment probability in transmission is further reduced.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of an audio denoising method for a bluetooth headset according to an embodiment of the present invention.
Detailed Description
In order to further describe the technical means and effects adopted by the invention to achieve the preset aim, the following detailed description refers to the specific implementation, structure, characteristics and effects of an audio denoising method for a bluetooth headset according to the present invention with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
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 invention belongs.
The following specifically describes a specific scheme of an audio denoising method for a bluetooth headset according to the present invention with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of an audio denoising method for a bluetooth headset according to an embodiment of the invention is shown, where the method includes:
step S1: acquiring an audio signal, and dividing an audio paragraph according to a trend item curve obtained by fitting the amplitude of the audio signal; and obtaining a fitting deviation value of each audio segment according to the audio signal and the trend item curve.
In the embodiment of the invention, the audio denoising method of the Bluetooth headset mainly aims at improving the quality of audio data in wireless transmission, the quality of audio signals depends on the size of sampling frequency to a certain extent, the higher the sampling frequency is, the higher the signal reduction degree is, but the transmission time of the audio data is correspondingly increased, and the longer the transmission time is, the higher the attachment rate of noise is, and the more dense the noise interference is. Therefore, the embodiment of the invention focuses on optimizing the coding sampling number of the audio data in wireless transmission, the wireless transmission has the defect of poor anti-interference capability, and the redundant sampling points of the audio file can increase the attachment probability of noise, so that the sampling points are selected from the lowest sampling point number, the self-adaptive increase is required according to the characteristics of each audio segment, the audio integrity is ensured, and simultaneously, the transmission time of the audio and the noise generated by electromagnetic interference in the transmission process are reduced to the greatest extent. Therefore, when analyzing the audio signal, firstly, the audio paragraphs are required to be obtained, then the demand characteristics of the sampling points are increased according to the fitting deviation value of each audio paragraph, so that the audio signal is obtained, the audio paragraphs are divided according to the trend item curve obtained by fitting the amplitude of the audio signal, and the fitting deviation value of each audio paragraph is obtained according to the audio signal and the trend item curve, and the method specifically comprises the following steps:
in the embodiment of the invention, a section of audio data is intercepted from the multimedia device, and because the nature of sound is a wave transmitted by vibration, and the sound can be considered as a series of analog signals when transmitted through an audio line or a microphone, the recording form of the audio signal is a time-series continuous waveform diagram, wherein the vertical axis of the waveform diagram is the amplitude of sound wave energy, the horizontal axis is time, and the number of sound waves in unit time is frequency. It should be noted that, the acquisition of the audio signal is a technical means well known to those skilled in the art, and will not be described herein.
The amplitude of the audio signal is used as a data point, a least square method is adopted for curve fitting to obtain a trend item curve, and the obtained trend item curve can be used for characterizing sound tones such as time sequence of sound in a section of audio, so that the smoothness of the trend item curve needs to be considered when a sampling point is selected. It should be noted that, the least square fitting is a technical means well known to those skilled in the art, and will not be described herein.
Since the trend term curve characterizes the sound characteristics, the audio paragraphs may be divided according to the trend term curve, so that each audio paragraph may be analyzed later to obtain the number of sampling points of each audio paragraph. Firstly, extremum in a trend item curve is obtained, a time sequence paragraph between every two adjacent extremum is divided into an audio paragraph, all audio paragraphs in an audio signal are obtained, and each audio paragraph has monotonicity. In the embodiment of the present invention, the extremum is obtained by deriving the trend term curve, and the point with the derivative being zero is taken as the extremum point, and it should be noted that the derivation and the extremum obtaining are technical means well known to those skilled in the art, and are not described herein.
More or less sound features may exist in each divided audio segment, and the fitting deviation value of the moment is obtained through the amplitude of the original audio signal and the corresponding value on the trend item curve, which specifically includes:
obtaining signal amplitude values at each moment in the audio signal, obtaining fitting values at each moment on a trend term curve, taking the difference value between the signal amplitude values corresponding to each moment and the fitting values as fitting deviation values, wherein the expression of the fitting deviation values is as follows:
in the method, in the process of the invention,
represented as a value of the fitting deviation,
represented as at the first
The amplitude of the time-of-day audio signal,
represented as the first on the trend term curve
Fitting values for time of day.
The difference between the trend item curve and the amplitude of the audio signal is obtained by using a difference method, when the difference is larger, the fitting deviation value is larger, the sound characteristic at the corresponding moment is obvious, meanwhile, for the whole audio paragraph, the fitting deviation value can reflect the discreteness of the audio data on the time sequence of the audio data, the discreteness reflects the distribution condition of the sound characteristic, when the data is larger, the sound characteristic in the audio paragraph is more, more sampling points are required to be set, and the discreteness is smaller, the distribution of the sound characteristic is less, and the number of the sampling points is also less.
Step S2: and obtaining a deviation value set consisting of fitting deviation values in each audio paragraph according to the distribution of the fitting deviation values, and obtaining the first sampling density of each audio paragraph according to the difference of the fitting deviation values in the deviation value set.
According to the step S1, the sound feature distribution situation in the audio paragraphs can be obtained by analyzing the fitting deviation values, more sampling points are added to the audio paragraphs with more sound features to ensure the integrity of the audio, and the audio paragraphs with fewer sound feature distributions do not need redundant sampling points to retain the signal features, so that when the fitting deviation values are analyzed, the variance of the fitting deviation values in each audio paragraph is analyzed, a deviation value set consisting of the fitting deviation values in each audio paragraph is obtained according to the distribution of the fitting deviation values, and the first sampling density of each audio paragraph is obtained according to the variance of the fitting deviation values in the deviation value set, which specifically includes:
according to the fluctuation characteristics of the audio signal, for the audio paragraph with the fitting deviation value, the audio paragraph has a position with a larger fitting deviation value, and the distribution of the fitting deviation value has the characteristic of Gaussian normalization, namely the amplitude position corresponding to the audio signal and the adjacent amplitude value of the audio signal are in a normal distribution shape with high middle and low two sides. Therefore, in order to better analyze the whole audio paragraph, the fitting deviation values in the audio paragraph are all arranged according to a normal distribution mode, that is, the fitting deviation values are sequentially decreased from the middle maximum value to the two ends, a deviation value set and a sequence number corresponding to each fitting deviation value in the deviation value set are obtained, and it is required to be noted that if the fitting deviation value is zero, it is noted that the time sequence point does not have obvious sound characteristics, and the fitting deviation value which is zero is screened out.
The overall analysis can be performed on each audio segment according to the deviation value set, when the difference of the fitting deviation values is larger, that is, the variation trend is larger, the audio segment needs more sampling points, so that the first sampling density of each audio segment is obtained according to the deviation value set, and the specific first sampling density is obtained by the steps of:
and obtaining the maximum fitting deviation value and the corresponding maximum sequence number in the deviation value set, comprehensively analyzing the two sides of the maximum fitting deviation value in the deviation value set according to the characteristic of normal distribution of the fitting deviation value, and comprehensively judging the trend change of the two sides. The minimum fitting deviation value at the left side of the maximum fitting deviation value is obtained and marked as a left minimum fitting deviation value, the serial number corresponding to the left minimum fitting deviation value is marked as a left minimum serial number, the minimum fitting deviation value at the right side of the maximum fitting deviation value is obtained and marked as a right minimum fitting deviation value, and the serial number corresponding to the right minimum fitting deviation value is marked as a right minimum serial number.
And respectively obtaining a deviation difference value and a sequence difference value of the fitting deviation values, and comprehensively analyzing the two sides to obtain the overall variation trend condition of the fitting deviation values, wherein the average value of the left minimum fitting deviation value and the right minimum fitting deviation value is taken as an average minimum deviation value, and the difference value of the maximum fitting deviation value and the average minimum deviation value is taken as the deviation difference value. Taking the absolute value of the difference between the maximum sequence number and the left minimum sequence number as a left sequence difference, taking the absolute value of the difference between the maximum sequence number and the right minimum sequence number as a right sequence difference, and taking the average value of the left sequence difference and the right sequence difference as a sequence difference value.
Taking the ratio of the deviation error value to the sequence error value as the first sampling density, in the embodiment of the invention, for the accuracy of subsequent calculation, the expression of the first sampling density is as follows:
in the method, in the process of the invention,
a first sampling density represented as an audio paragraph;
represented as the largest fitting deviation value in the set of deviation values;
the maximum sequence number corresponding to the maximum fitting deviation value is expressed;
the minimum fitting deviation value, expressed as the left of the maximum fitting deviation value, i.e., the left minimum fitting deviation value;
the left minimum sequence number is represented as the corresponding left minimum fitting deviation value;
represented as the left side of the maximum fitted deviation value in the set of deviation values;
the minimum fitting deviation value to the right of the maximum fitting deviation value, i.e., the right minimum fitting deviation value;
the right minimum sequence number corresponding to the right minimum fitting deviation value is represented;
represented to the right of the maximum fitting deviation value in the set of deviation values.
Comprehensively analyzing the variation trend at two sides of the maximum fitting deviation value in normal distribution by utilizing the form of the ratio,
represented as the average minimum deviation value obtained by averaging the left and right minimum fitting deviation values,
expressed as a deviation error value, the two sides of the maximum fitting deviation value are combined to obtain the error of a deviation value set,
represented as the left sequence difference,
represented as the right sequence difference,
and the sequence difference value is obtained by averaging the left sequence difference and the right sequence difference, and the sequence difference which corresponds to the deviation value set and reflects the change trend is obtained on two sides of the maximum fitting deviation value. The ratio of the deviation value to the sequence difference value represents the variation trend of the fitting deviation value of each audio paragraph, namely the first sampling density, and when the variation trend of the fitting deviation value is larger, the larger the fitting deviation value is, the more sound characteristic distribution of the corresponding audio paragraph is further described, and at the moment, the sampling density needs to be increased, namely the number of sampling points is increased.
Step S3: obtaining probability density distribution of each type of fitting deviation values according to the type density distribution conditions of all fitting deviation values in the audio signal; and taking any type of fitting deviation values in the deviation value set of any one audio paragraph as reference fitting deviation values, acquiring type influence values of the reference fitting deviation values through the duty ratio of the reference fitting deviation values and the corresponding probability density distribution, and acquiring the second sampling density of the audio paragraph according to the type influence values of the fitting deviation values of all types.
In step S2, the difference analysis of the fitting deviation value of each audio segment is completed, and further, the probability density distribution of the fitting deviation value of different types in each audio segment in the whole audio signal is analyzed. In the audio signal, because the probability of occurrence of the fitting deviation values of different types is different, and the distribution probability influence of the fitting deviation values of different types in each audio paragraph on the fitting deviation values of the whole audio signal is also different, the probability density of the fitting deviation values of different types is analyzed, and the second sampling density is obtained according to the influence condition of the fitting deviation values of different types in each audio paragraph on the whole audio signal. Therefore, firstly, probability density distribution conditions corresponding to fitting deviation values of all types in an audio signal are obtained, and probability density distribution of each type of fitting deviation values is obtained according to the type density distribution conditions of all fitting deviation values in the audio signal, and specifically comprises the following steps:
preferably, the type and the number of all fitting deviation values in the audio signal are counted to obtain a distribution probability histogram of the fitting deviation values, the type distribution probability of each type of fitting deviation values is obtained according to the distribution probability histogram, and each type distribution probability represents the distribution probability condition of the fitting deviation values of the corresponding type. In order to further more accurately represent the influence of each type of fitting deviation value in each audio paragraph, the result obtained by taking each type of distribution probability as an argument of a gaussian probability density function is taken as probability density distribution of the corresponding type of fitting deviation value, the probability density distribution value reflects the probability density of the type of distribution probability of the corresponding type of fitting deviation value in the gaussian probability density function, and in the embodiment of the invention, the probability density distribution formula is as follows in consideration of convenience of subsequent calculation:
in the method, in the process of the invention,
denoted as the first
The probability density distribution of the type distribution probability of the class fitting deviation value in the Gaussian probability density function is as follows
，
Denoted as the first
The probability of the type distribution of the class fitting bias values,
the standard deviation of the type distribution probability expressed as the fitting deviation value,
represented as an average of the type distribution probabilities of the fitting deviation values,
represented by the rate of the circumference of a circle,
represented as an exponential function with a base of natural constant. It should be noted that, the application of the gaussian probability density function formula is a technical means well known to those skilled in the art, so the meaning of the specific formula is not repeated.
After probability density distribution of fitting deviation values of all types is obtained, each audio paragraph is analyzed, different numbers of sampling points are selected for the audio paragraphs according to different influences of the fitting deviation values of each type on distribution probability in the whole audio signal, therefore, any type of fitting deviation value in a deviation value set of any audio paragraph is taken as a reference fitting deviation value, the type influence value of the reference fitting deviation value is obtained through the duty ratio of the reference fitting deviation value and the corresponding probability density distribution, and the second sampling density of the audio paragraph is obtained according to the type influence value of the fitting deviation values of all types, and the method specifically comprises the following steps:
because each type of fitting deviation value in each audio paragraph needs to be analyzed, any type of fitting deviation value in the deviation value set of any audio paragraph is taken as a reference fitting deviation value, the probability density distribution corresponding to the reference fitting deviation value is taken as a first reference value, and the first reference value reflects the probability that the distribution probability of the reference fitting deviation value in the audio paragraph is the probability of the type distribution probability, namely the type distribution probability of the reference fitting deviation value is the expected value of the Gaussian probability density function.
Taking the ratio of the number of the reference fitting deviation values in the audio paragraph to the total number of the reference fitting deviation values in the audio signal as a number duty ratio, wherein the number duty ratio reflects the duty ratio condition of the reference fitting deviation values in the whole audio signal for the reference fitting deviation values in the audio paragraph, when the duty ratio of the reference fitting deviation values in the audio paragraph is larger, the influence degree of the reference fitting deviation values of the audio paragraph is larger, the number duty ratio is multiplied by a first reference value, and the type influence value of the reference fitting deviation values is obtained, and in the embodiment of the invention, the type influence value expression is given in consideration of the accuracy of subsequent calculation:
in the method, in the process of the invention,
denoted as the first
The first audio paragraph
The type of class fit bias value affects the value,
denoted as the first
The number of audio paragraphs,
denoted as the first
Class fitting deviation values i.e. reference fitting deviation values,
denoted as the first
The first audio paragraph
The number of class fit bias values,
denoted as the first
The number of class fit offset values in the overall audio signal,
denoted as the first
The probability density distribution of the type distribution probability of the class fitting deviation value in the Gaussian probability density function is as follows
，
Denoted as the first
The probability of the type distribution of the class fitting bias values,
denoted as pair I
The type distribution probability of the class fitting deviation value takes a first reference value of the fixed integral result, namely the first reference value
The type distribution probability of the class fitting bias values is the expected value in a gaussian probability density function.
Comprehensively analyzing the quantity duty ratio and the first reference value in the form of product,
denoted as the first
The first audio paragraph
Class fitting deviation value in the whole audio signal
The larger the number of the class fitting deviation values, which indicates that the influence of the reference fitting deviation values in the audio paragraph on the whole audio signal is, so that the type influence value is larger; when the first reference valueThe larger is illustrated in the first
The more likely the distribution probability of the reference fitting deviation value in the audio paragraph is the type distribution probability, the reference fitting deviation value distribution of the audio paragraph is close to the distribution of the reference fitting deviation value in the whole audio signal, which indicates the first time
The effect of the reference fitting offset values in the audio paragraphs is larger, and therefore the greater the type effect value.
The type influence value characterizes the influence of the reference fitting deviation value in the audio paragraph on the distribution probability of the reference fitting deviation value in the whole audio signal, and when the type influence value is larger, the influence of the reference fitting deviation value in the audio paragraph on the reference fitting deviation value in the whole audio signal is larger, and sampling points are required to be added for the audio paragraph, so that the whole audio signal is more complete, and the reserved information is more accurate.
Obtaining type influence values of fitting deviation values of all types in an audio paragraph, normalizing the type influence values of the fitting deviation values of each type in the audio paragraph to obtain corresponding normalized type influence values, taking the addition result of the normalized type influence values as a second sampling density of the audio paragraph, wherein the second sampling density synthesizes the influence of the fitting deviation values of all types in each audio paragraph, and sampling points are required to be added to the audio paragraph with large influence so as to ensure the integrity of the whole audio signal.
In the method, in the process of the invention,
denoted as the first
Second sampling density of audio paragraphs，
Denoted as the first
The deviation value is fitted in a similar way,
expressed as the total number of fitting offset value types,
denoted as the first
The first audio paragraph
The type of class fit bias value affects the value,
the hyperbolic tangent function is shown, and it should be noted that the hyperbolic tangent function is a technical means well known to those skilled in the art, and is not described herein.
Fitting deviation values of all classes of audio paragraphs are comprehensively analyzed in an accumulating mode, and the method is utilized
The hyperbolic tangent function will be
Normalizing the type influence values of the class fitting deviation values, adopting normalization processing to limit the value range of the numerical values, adding the type influence values of all the normalized class fitting deviation values, taking the obtained accumulated value as the second sampling density of the audio paragraphs, and when the accumulated value is larger, indicating that the distribution influence of each type of fitting deviation value in the whole audio signal is larger in the corresponding audio paragraphs, considering that the influence of the audio paragraphs on the whole audio signal is extremely large, and the more important the audio paragraphs are needed to be paid attention to ensure the integrity of the whole audio signal.
Step S4: obtaining a sampling density coefficient according to the first sampling density and the second sampling density, and obtaining the number of sampling points of each audio paragraph according to the duty ratio of each audio paragraph in the audio signal and the sampling density coefficient; and sampling and denoising the audio according to the number of the sampling points.
According to the step S2 and the step S3, the difference of fitting deviation values in each audio paragraph and the influence of different fitting deviation values in each audio paragraph on the whole audio signal are respectively analyzed, the first sampling density and the second sampling density of each audio paragraph are respectively obtained, and finally, the final sampling density coefficient is obtained through integration. The final sampling density coefficient can reflect the degree of the number of sampling points required to be increased for each audio paragraph, the theoretical minimum number of sampling points of each audio paragraph is calculated and obtained, and the number of sampling points is adaptively increased according to the sampling density coefficient based on the minimum number of sampling points, so that the integrity of an audio signal is ensured. Thus, the sampling density coefficient is obtained according to the first sampling density and the second sampling density, and the sampling point number of each audio segment is obtained according to the duty ratio of each audio segment in the audio signal and the sampling density coefficient, which specifically comprises the following steps:
taking an average value of the obtained first sampling density and the second sampling density of each audio segment as a final sampling density coefficient, wherein the expression of the sampling density coefficient is as follows:
in the method, in the process of the invention,
represented as the sampling density coefficient of an audio segment,
represented as a first sampling density of the audio paragraphs,
represented as a second sampling density of the audio paragraphs.
The first sampling density and the second sampling density are integrated in a mean value mode, and when the first sampling density and the second sampling density are larger, the audio paragraph is required to be added with more sampling points to ensure the integrity of an audio signal, so that the sampling density coefficient is larger.
When the audio signal is stored and transmitted, the encoding process needs to be carried out firstly, but the general encoding sampling rate cannot be lower than half of the audio signal, the higher the sampling rate, the more the acquisition number per second is, the closer the sampling points are, the smoother the obtained audio curve is, and the more natural the transition is. But a higher sampling rate also means a larger storage of the audio file, a longer transmission time and a higher probability of noise interference. Thus, when sampling points are selected, the lowest number of sampling points of the whole audio signal, i.e. half the number of waves in the whole original audio signal, is obtained first.
And (3) analyzing any audio paragraph, multiplying the minimum sampling point number by the length ratio of the audio paragraph in the audio signal to obtain the paragraph minimum sampling point number of the audio paragraph, wherein the paragraph minimum sampling point number of each audio paragraph represents the put-in number of theoretical sampling points in each audio paragraph, and increasing the sampling point number according to the corresponding sampling density coefficient on the basis of the paragraph minimum sampling point number. According to the method, sampling distribution weight is obtained according to the sampling density coefficient, the lowest sampling point number of the paragraphs is multiplied with the sampling distribution weight, and the sampling point number of the audio paragraphs is obtained, and in the embodiment of the invention, the obtaining expression of the sampling point number of the audio paragraphs is as follows in consideration of calculation accuracy:
in the method, in the process of the invention,
denoted as the first
The number of sample points for the individual audio segments,
expressed as wholeThe length of the individual audio signals is chosen,
expressed as the number of original signal waves of the whole audio signal,
denoted as the first
The length of the individual audio passages,
denoted as the first
The sampling density coefficients of the individual audio segments,
represented as the lowest number of sampling points,
denoted as the first
The length of the individual audio segments is the duty cycle of the length of the entire audio signal,
represented as sample distribution weights.
The number of the lowest sampling points of each audio paragraph is adaptively increased according to different sampling distribution weights, the number of sampling points which are adaptively increased on the theoretical sampling point throwing number is finally obtained,
expressed as the lowest sampling point of the paragraph, i.e. the theoretical sampling point placement number of the audio paragraph,
represented as sample distribution weights, in which
The weight of the number of sampling points which is required to be increased according to the lowest number of sampling points of the paragraphs is increased according to different influences of each audio paragraph, and it is required to be noted that the data types corresponding to the number of sampling points are integers, so that the number of sampling points obtained through final calculation is rounded downwards to obtain the number of real sampling points of each audio paragraph.
And sampling the audio signal according to the number of the sampling points and denoising the audio signal to finish denoising. In the embodiment of the invention, the audio data obtained by sampling is quantized, coded and transmitted by adopting a PCM coding method. It should be noted that PCM encoding is a technical means well known to those skilled in the art, and will not be described herein.
The self-adaptive sampling point method considers the characteristics of the signal and the integrity of the falling signals of different audio frequency segments, and based on the number of the lowest standard sampling points, the self-adaptive increase is carried out according to the requirements of the falling signals of different audio frequency segments, so that the signal integrity is ensured, the audio coding efficiency is higher, unnecessary sampling points are reduced, the number of the sampling points is reduced to the greatest extent, the transmission interaction time is reduced, the noise attachment rate during transmission is reduced, and the quality and the efficiency of audio denoising are improved.
In summary, according to the invention, an audio signal is obtained, a trend term curve is obtained according to amplitude fitting of the audio signal, a fitting deviation value is obtained according to the audio signal and the trend term curve, a first sampling density of each audio paragraph is obtained by analyzing the difference of the fitting deviation value for each audio paragraph, a second sampling density of each audio paragraph is obtained by analyzing probability density distribution of different types of fitting deviation values in the audio paragraphs in the whole audio signal for each audio paragraph, a sampling density coefficient is obtained according to the first sampling density and the second sampling density, the number of sampling points of each audio paragraph is obtained according to the duty ratio and the sampling density coefficient of each audio paragraph in the audio signal, and audio is sampled and denoised according to the number of sampling points. According to the invention, through analyzing the audio data, the characteristics of the signal and the integrity of the falling signals of different audio segments are considered, and the self-adaptive increase is carried out according to the requirements of the falling signals of different audio segments based on the number of the lowest standard sampling points.
It should be noted that: the sequence of the embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.
Claims (10)
1. An audio denoising method for a bluetooth headset, the method comprising:
acquiring an audio signal, and dividing an audio paragraph according to a trend item curve obtained by fitting the amplitude of the audio signal; obtaining a fitting deviation value of each audio segment according to the audio signal and the trend item curve;
obtaining a deviation value set consisting of fitting deviation values in each audio paragraph according to the distribution of the fitting deviation values, and obtaining a first sampling density of each audio paragraph according to the difference of the fitting deviation values in the deviation value set;
obtaining probability density distribution of each type of fitting deviation values according to the type density distribution conditions of all fitting deviation values in the audio signal; taking any type of fitting deviation values in the deviation value set of any one audio paragraph as reference fitting deviation values, acquiring type influence values of the reference fitting deviation values through the duty ratio of the reference fitting deviation values and the corresponding probability density distribution, and acquiring second sampling density of the audio paragraph according to the type influence values of the fitting deviation values of all types;
obtaining a sampling density coefficient according to the first sampling density and the second sampling density, and obtaining the number of sampling points of each audio paragraph according to the duty ratio of each audio paragraph in the audio signal and the sampling density coefficient; and sampling and denoising the audio according to the number of the sampling points.
2. The method for audio denoising of a bluetooth headset according to claim 1, wherein the obtaining of the set of offset values comprises:
and arranging the fitting deviation values in the audio paragraphs according to a normal distribution mode to obtain a deviation value set and a serial number of each fitting deviation value in the deviation value set.
3. The audio denoising method for a bluetooth headset according to claim 2, wherein the acquisition of the first sampling density comprises:
obtaining the maximum fitting deviation value and the corresponding maximum sequence number in the deviation value set; in the deviation value set, the minimum fitting deviation value at the left side of the maximum fitting deviation value is obtained and marked as a left minimum fitting deviation value, the serial number corresponding to the left minimum fitting deviation value is marked as a left minimum serial number, the minimum fitting deviation value at the right side of the maximum fitting deviation value is obtained and marked as a right minimum fitting deviation value, and the serial number corresponding to the right minimum fitting deviation value is marked as a right minimum serial number;
taking the average value of the left minimum fitting deviation value and the right minimum fitting deviation value as an average minimum deviation value, and taking the difference value of the maximum fitting deviation value and the average minimum deviation value as a deviation difference value; taking the absolute value of the difference between the maximum sequence number and the left minimum sequence number as a left sequence difference, taking the absolute value of the difference between the maximum sequence number and the right minimum sequence number as a right sequence difference, and taking the average value of the left sequence difference and the right sequence difference as a sequence difference value;
the ratio of the deviation error value to the sequence error value is taken as a first sampling density.
4. The audio denoising method for a bluetooth headset according to claim 1, wherein the acquiring of the probability density distribution comprises:
counting the types and the quantity of all fitting deviation values in the audio signal to obtain a distribution probability histogram of the fitting deviation values, and obtaining the type distribution probability of each type of fitting deviation values according to the distribution probability histogram; the result obtained by taking each type of distribution probability as an independent variable of the Gaussian probability density function is taken as probability density distribution of the corresponding class fitting deviation value.
5. The audio denoising method for a bluetooth headset according to claim 1, wherein the acquiring of the type influence value comprises:
taking the result of the fixed integral of probability density distribution corresponding to the reference fitting deviation value as a first reference value;
and taking the ratio of the number of the reference fitting deviation values in the audio paragraph to the total number of the reference fitting deviation values in the audio signal as a number duty ratio, and multiplying the number duty ratio by the first reference value to obtain the type influence value of the reference fitting deviation values.
6. The audio denoising method for a bluetooth headset according to claim 1, wherein the obtaining of the second sampling density comprises:
normalizing the type influence value of each type of fitting deviation value in the audio paragraph to obtain a corresponding normalized type influence value, and taking the addition result of the normalized type influence values as the second sampling density of the audio paragraph.
7. The audio denoising method for a bluetooth headset according to claim 1, wherein the acquisition of the sampling density coefficient comprises:
taking the average value of the first sampling density and the second sampling density as a sampling density coefficient.
8. The method for audio denoising of a bluetooth headset according to claim 1, wherein the obtaining the number of sampling points for each audio segment comprises:
obtaining the lowest sampling point number of the audio signal, and multiplying the lowest sampling point number by the length ratio of the audio paragraph in the audio signal for any audio paragraph to obtain the lowest sampling point number of the audio paragraph; and obtaining sampling distribution weights according to the sampling density coefficients, multiplying the lowest sampling point number of the paragraphs by the sampling distribution weights, and obtaining the sampling point number of the audio paragraphs.
9. The audio denoising method for a bluetooth headset according to claim 1, wherein the acquiring of the audio segment comprises:
taking the amplitude of the audio signal as a data point, and adopting a least square method to perform curve fitting to obtain a trend item curve; and obtaining extreme values in the trend item curve, dividing a time sequence paragraph between every two adjacent extreme values into an audio paragraph, and obtaining all the audio paragraphs in the audio signal.
10. The audio denoising method for a bluetooth headset according to claim 1, wherein the obtaining of the fitting deviation value comprises:
and obtaining the signal amplitude of each moment in the audio signal, obtaining the fitting value of each moment on the trend term curve, and taking the difference value of the signal amplitude and the fitting value corresponding to each moment as the fitting deviation value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310043255.5A CN115798502B (en) | 2023-01-29 | 2023-01-29 | Audio denoising method for Bluetooth headset |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310043255.5A CN115798502B (en) | 2023-01-29 | 2023-01-29 | Audio denoising method for Bluetooth headset |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115798502A CN115798502A (en) | 2023-03-14 |
| CN115798502B true CN115798502B (en) | 2023-04-25 |
Family
ID=85429057
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310043255.5A Active CN115798502B (en) | 2023-01-29 | 2023-01-29 | Audio denoising method for Bluetooth headset |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115798502B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116405823B (en) * | 2023-06-01 | 2023-08-29 | 深圳市匠心原创科技有限公司 | Intelligent audio denoising enhancement method for bone conduction earphone |
| CN116991199B (en) * | 2023-09-25 | 2024-01-02 | 深圳扬兴科技有限公司 | Clock signal intelligent noise reduction method for clock chip |
| CN117357094A (en) * | 2023-11-15 | 2024-01-09 | 读取(北京)科技有限公司 | Respiratory state verification system using sound intensity and carbon dioxide detection |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0386820A1 (en) * | 1989-03-03 | 1990-09-12 | Koninklijke Philips Electronics N.V. | Method and arrangement for probabilistic pitch measuring |
| CN105117537A (en) * | 2015-08-13 | 2015-12-02 | 电子科技大学 | Weight comparison based resampling method for particle filter system |
| CN109102818A (en) * | 2018-08-29 | 2018-12-28 | 昆明理工大学 | A kind of denoising audio sample algorithm based on signal frequency probability density function profiles |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4517045B2 (en) * | 2005-04-01 | 2010-08-04 | 独立行政法人産業技術総合研究所 | Pitch estimation method and apparatus, and pitch estimation program |
| US9787341B2 (en) * | 2012-11-21 | 2017-10-10 | Alarm.Com Incorporated | Density function centric signal processing |
| CN109891504A (en) * | 2016-10-07 | 2019-06-14 | 索尼公司 | Information processing equipment and method and program |
-
2023
- 2023-01-29 CN CN202310043255.5A patent/CN115798502B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0386820A1 (en) * | 1989-03-03 | 1990-09-12 | Koninklijke Philips Electronics N.V. | Method and arrangement for probabilistic pitch measuring |
| CN105117537A (en) * | 2015-08-13 | 2015-12-02 | 电子科技大学 | Weight comparison based resampling method for particle filter system |
| CN109102818A (en) * | 2018-08-29 | 2018-12-28 | 昆明理工大学 | A kind of denoising audio sample algorithm based on signal frequency probability density function profiles |
Non-Patent Citations (2)
| Title |
|---|
| 汤显峰 ; 葛泉波 ; .基于子带无迹粒子滤波的语音增强算法.自动化与仪器仪表.2011,(第04期),147-154. * |
| 郭小华 ; 陈文岳 ; .基于压缩感知的资源节约型音频信号采集方法.农业工程学报.2013,(第16期),174-181. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115798502A (en) | 2023-03-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115798502B (en) | Audio denoising method for Bluetooth headset | |
| CN102918588B (en) | A spatial audio processor and a method for providing spatial parameters based on an acoustic input signal | |
| JP5089772B2 (en) | Apparatus and method for detecting voice activity | |
| CN104347067B (en) | Audio signal classification method and device | |
| CN1284133C (en) | Distributed voice recognition system using acoustic feature vector modification | |
| CN108538310B (en) | Voice endpoint detection method based on long-time signal power spectrum change | |
| CN102959625B9 (en) | Method and apparatus for adaptively detecting voice activity in input audio signal | |
| CN108447495B (en) | Deep learning voice enhancement method based on comprehensive feature set | |
| CN103117067B (en) | Voice endpoint detection method under low signal-to-noise ratio | |
| US8682664B2 (en) | Method and device for audio signal classification using tonal characteristic parameters and spectral tilt characteristic parameters | |
| KR20070038118A (en) | Device and method for robustry classifying audio signals, method for establishing and operating audio signal database and a computer program | |
| CN103733257A (en) | Audio-encoding method and apparatus, audio-decoding method and apparatus, recording medium thereof, and multimedia device employing same | |
| CN102549659A (en) | Suppressing noise in an audio signal | |
| CN103325384A (en) | Harmonicity estimation, audio classification, pitch definition and noise estimation | |
| CN103247298B (en) | A kind of sensitivity correction method and audio frequency apparatus | |
| CN103544957B (en) | Method and device for bit distribution of sound signal | |
| GB2585086A (en) | Pre-processing for automatic speech recognition | |
| CN116364109A (en) | Speech enhancement network signal-to-noise ratio estimator and loss optimization method | |
| JP3590342B2 (en) | Signal encoding method and apparatus, and recording medium recording signal encoding program | |
| CN111341351B (en) | Voice activity detection method, device and storage medium based on self-attention mechanism | |
| US20230245668A1 (en) | Neural network-based audio packet loss restoration method and apparatus, and system | |
| CN110007342B (en) | Time-frequency domain direct pickup first arrival method and system for seismic signals with low signal-to-noise ratio | |
| CN112992157A (en) | Neural network noisy line identification method based on residual error and batch normalization | |
| CN113113049A (en) | Voice activity detection method combined with voice enhancement | |
| CN113327589A (en) | Voice activity detection method based on attitude sensor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |