1. A quality evaluation method based on voice data is characterized by comprising the following steps:

step one, voice preprocessing, namely preprocessing voice data so as to detect the voice data;

step two, data analysis, namely performing data analysis on the voice data which is preprocessed to obtain a quality analysis coefficient of the voice data;

thirdly, quality evaluation, namely constructing a quality evaluation model and carrying out quality evaluation on the voice data;

the voice data preprocessing in the first step comprises the following specific steps:

step S1: acquiring a section of 20-second voice, marking the voice as test voice, acquiring a section of voice with qualified quality evaluation, marking the voice as standard voice, continuously playing the test voice for one minute, acquiring a voice sequence interval corresponding to a monitoring point in the test voice within one minute by taking a time point of each second as the monitoring point, wherein the voice sequence interval is represented as the interval duration of two bytes passing through the monitoring point when the bytes in the voice are played, and then acquiring a voice sequence interval corresponding to the monitoring point in the standard voice within one minute;

step S2: comparing the voice sequence interval in the monitored voice with the voice sequence interval in the standard voice, if the voice sequence interval in the monitored voice is larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is unqualified, marking the tested voice sequence interval as abnormal voice of the sequence interval, if the voice sequence interval in the monitored voice is not larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is qualified, and entering the step S3;

step S3: performing frame division on monitoring voice, dividing the monitoring voice into a plurality of voice frames, marking the voice frames as i, i is 1, 2, … …, n and n are positive integers, acquiring the audio wavelength and frequency of the voice frames, respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi, and then respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi through a formula Xi is beta (Bi x a1+ Pi x a2)²Acquiring a parameter analysis coefficient Xi of the voice frame, wherein a1 and a2 are both proportional coefficients, and beta is an error correction factor and takes the value of 1.36;

step S4: constructing a parameter analysis coefficient set for a voice frame in the test voice, comparing subsets in the parameter analysis coefficient set with a parameter analysis coefficient threshold, if the subsets are more than or equal to the parameter analysis coefficient threshold, marking the corresponding subsets as qualified subsets, and if the subsets are less than the parameter analysis coefficient threshold, marking the corresponding subsets as unqualified subsets; if the number of qualified subsets in the parameter analysis coefficient set is larger than that of unqualified subsets, acquiring the average value of the parameter analysis coefficient set and marking the average value as a parameter analysis coefficient RS of the tested voice, otherwise, marking the corresponding tested voice as unqualified-parameter voice.

2. The quality evaluation method based on voice data according to claim 1, wherein the specific steps of obtaining the quality analysis coefficient of the voice data in the second step are as follows:

step SS 1: marking the test voice as o, o is 1, 2, … …, m is a positive integer, acquiring the audio signal-to-noise ratio of each monitoring point of the test voice, and marking the audio signal-to-noise ratio of each monitoring point of the test voice as XZo;

step SS 2: acquiring an audio cycle of each monitoring point of the test voice, and marking the audio cycle of each monitoring point of the test voice as ZQo;

step SS 3: acquiring the audio wavelength of each monitoring point of the test voice, and marking the audio wavelength of each monitoring point of the test voice as BCo;

step SS 4: by the formulaAcquiring a quality analysis coefficient FXo of the test voice, wherein s1, s2 and s3 are all proportionality coefficients, s1 is more than s2 is more than s3 is more than 0, and e is a natural constant;

step SS 5: the quality analysis coefficient FXo of the test speech is compared to a quality analysis coefficient threshold.

3. The quality evaluation method based on the voice data according to claim 1, wherein the quality evaluation of the voice data is performed by constructing a quality evaluation model in the third step, and the specifically constructing and evaluating steps are as follows:

step T1: obtaining a parameter analysis coefficient RS of the test voice and a quality analysis coefficient FXo corresponding to the quality qualified test voice, and then constructing a quality evaluation model, wherein the quality evaluation model formula is PJ ═ alpha (e)^RS×b1+e^FXo×b2)×t]^3.65Wherein b1 and b2 are proportionality coefficients, b1 > b2 > 0, e is a natural constant, and t is an independent variable timePJ is a quality evaluation coefficient;

step T2: establishing a rectangular coordinate system by taking the independent variable time as an X axis and the quality evaluation coefficient as a Y axis, taking the integral point moment as a point on the X axis, marking the inner point of the rectangular coordinate system corresponding to the integral point moment on the X axis, and connecting the inner points in the rectangular coordinate system to construct a quality evaluation curve of the tested voice;

step T3: analyzing the quality evaluation curve, analyzing the corresponding integral point time in the total duration of the tested voice, obtaining the slope of the quality evaluation curve between two adjacent integral point times, and comparing the corresponding slope of the quality evaluation curve:

if the slope of the corresponding quality evaluation curve is less than 40 degrees, judging that the test voice corresponding to the time of the adjacent integral points is normal; and if the slope of the corresponding quality evaluation curve is less than 65 degrees and less than 40 degrees, judging that the testing voice corresponding to the adjacent integral point time is abnormal, marking the testing voice corresponding to the adjacent integral point time as abnormal voice, generating a voice rectification signal and sending the voice rectification signal to the mobile phone terminal of the manager.

Background

The speech signal is inevitably distorted after being transmitted and processed, and is especially accompanied by the appearance of various speech algorithms, such as speech coding, speech enhancement, speech synthesis, channel transmission, etc., which meet the requirements of some applications, but inevitably cause damage to speech, and the distortion category caused by the algorithms cannot be considered, so that the influence on the speech quality is unpredictable. The voice quality is an important index for detecting the performance of voice equipment and algorithms, so that how to evaluate the voice quality is very important.

However, in the prior art, the test voice cannot be preprocessed, so that the working intensity is increased, the working efficiency of voice quality evaluation is reduced, and meanwhile, the quality of the voice cannot be easily obtained, so that the working intensity of voice test is reduced.

Disclosure of Invention

The invention aims to provide a quality evaluation method based on voice data, which comprises the steps of obtaining a section of voice of 20 seconds, obtaining a section of voice qualified in quality evaluation, marking the voice as standard voice, continuously playing test voice for one minute, obtaining a voice sequence interval corresponding to a monitoring point in the test voice within one minute by taking a time point of each second as the monitoring point, wherein the voice sequence interval is represented as the interval duration of two bytes passing through the monitoring point when bytes in the voice are played, and then obtaining a voice sequence interval corresponding to the monitoring point in the standard voice within one minute; comparing the voice sequence interval in the monitoring voice with the voice sequence interval in the standard voice, performing frame division on the monitoring voice, dividing the monitoring voice into a plurality of voice frames, acquiring the audio wavelength and frequency of the voice frames, then acquiring the parameter analysis coefficient X i of the voice frames through a formula, constructing a parameter analysis coefficient set for the voice frames in the testing voice, and comparing the subset in the parameter analysis coefficient set with a parameter analysis coefficient threshold; the test voice is preprocessed, so that the qualification rate of the test voice is improved, the quality evaluation of unqualified voice is prevented, the working intensity is increased, and the working efficiency of voice quality evaluation is reduced;

the purpose of the invention can be realized by the following technical scheme:

a quality evaluation method based on voice data comprises the following steps:

step one, voice preprocessing, namely preprocessing voice data so as to detect the voice data;

step two, data analysis, namely performing data analysis on the voice data which is preprocessed to obtain a quality analysis coefficient of the voice data;

thirdly, quality evaluation, namely constructing a quality evaluation model and carrying out quality evaluation on the voice data;

the voice data preprocessing in the first step comprises the following specific steps:

step S1: acquiring a section of 20-second voice, marking the voice as test voice, acquiring a section of voice with qualified quality evaluation, marking the voice as standard voice, continuously playing the test voice for one minute, acquiring a voice sequence interval corresponding to a monitoring point in the test voice within one minute by taking a time point of each second as the monitoring point, wherein the voice sequence interval is represented as the interval duration of two bytes passing through the monitoring point when the bytes in the voice are played, and then acquiring a voice sequence interval corresponding to the monitoring point in the standard voice within one minute;

step S2: comparing the voice sequence interval in the monitored voice with the voice sequence interval in the standard voice, if the voice sequence interval in the monitored voice is larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is unqualified, marking the tested voice sequence interval as abnormal voice of the sequence interval, if the voice sequence interval in the monitored voice is not larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is qualified, and entering the step S3;

step S3: performing frame division on monitoring voice, dividing the monitoring voice into a plurality of voice frames, marking the voice frames as i, i is 1, 2, … …, n and n are positive integers, acquiring the audio wavelength and frequency of the voice frames, respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi, and then respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi through a formula Xi is beta (Bi x a1+ Pi x a2)²Acquiring a parameter analysis coefficient Xi of the voice frame, wherein a1 and a2 are both proportional coefficients, and beta is an error correction factor and takes the value of 1.36;

step S4: constructing a parameter analysis coefficient set for a voice frame in the test voice, comparing subsets in the parameter analysis coefficient set with a parameter analysis coefficient threshold, if the subsets are more than or equal to the parameter analysis coefficient threshold, marking the corresponding subsets as qualified subsets, and if the subsets are less than the parameter analysis coefficient threshold, marking the corresponding subsets as unqualified subsets; if the number of qualified subsets in the parameter analysis coefficient set is larger than that of unqualified subsets, acquiring the average value of the parameter analysis coefficient set and marking the average value as a parameter analysis coefficient RS of the test voice, otherwise, marking the corresponding test voice as unqualified-parameter voice; the test voice is preprocessed, so that the qualification rate of the test voice is improved, the quality evaluation of unqualified voices is prevented, the working intensity is increased, and the working efficiency of voice quality evaluation is reduced.

Further, the specific steps of obtaining the quality analysis coefficient of the voice data in the step two are as follows:

step SS 1: marking the test voice as o, o is 1, 2, … …, m is a positive integer, acquiring the audio signal-to-noise ratio of each monitoring point of the test voice, and marking the audio signal-to-noise ratio of each monitoring point of the test voice as XZo;

step SS 2: acquiring an audio cycle of each monitoring point of the test voice, and marking the audio cycle of each monitoring point of the test voice as ZQo;

step SS 3: acquiring the audio wavelength of each monitoring point of the test voice, and marking the audio wavelength of each monitoring point of the test voice as BCo;

step SS 4: by the formulaAcquiring a quality analysis coefficient FXo of the test voice, wherein s1, s2 and s3 are all proportionality coefficients, s1 is more than s2 is more than s3 is more than 0, and e is a natural constant;

step SS 5: the quality analysis coefficient FXo of the test speech is compared to a quality analysis coefficient threshold:

if the quality analysis coefficient FXo of the test voice is larger than or equal to the quality analysis coefficient threshold, judging that the quality of the test voice is qualified, generating a quality qualified signal and marking the corresponding test voice as a quality qualified test voice;

and if the quality analysis coefficient FXo of the test voice is less than the quality analysis coefficient threshold value, judging that the quality of the test voice is unqualified, generating a quality unqualified signal and marking the corresponding test voice as the quality unqualified test voice.

Further, in the third step, a quality evaluation model is constructed for quality evaluation of the voice data, and the specific construction and evaluation steps are as follows:

step T1: obtaining a parameter analysis coefficient RS of the test voice and a quality analysis coefficient FXo corresponding to the quality qualified test voice, and then constructing a quality evaluation model, wherein the quality evaluation model formula is PJ ═ alpha (e)^RS×b1+e^FXo×b2)×t]^3.65Wherein, b1 and b2 are proportionality coefficients, b1 is larger than b2 is larger than 0, e is a natural constant, t is independent variable time, and PJ is a quality evaluation coefficient;

step T2: establishing a rectangular coordinate system by taking the independent variable time as an X axis and the quality evaluation coefficient as a Y axis, taking the integral point moment as a point on the X axis, marking the inner point of the rectangular coordinate system corresponding to the integral point moment on the X axis, and connecting the inner points in the rectangular coordinate system to construct a quality evaluation curve of the tested voice;

step T3: analyzing the quality evaluation curve, analyzing the corresponding integral point time in the total duration of the tested voice, obtaining the slope of the quality evaluation curve between two adjacent integral point times, and comparing the corresponding slope of the quality evaluation curve:

if the slope of the corresponding quality evaluation curve is less than 40 degrees, judging that the test voice corresponding to the time of the adjacent integral points is normal; if the slope of the corresponding quality evaluation curve is less than 65 degrees and less than 40 degrees, judging that the testing voice corresponding to the adjacent integral point time is abnormal, marking the testing voice corresponding to the adjacent integral point time as abnormal voice, generating a voice rectification signal and sending the voice rectification signal to a mobile phone terminal of a manager; quality evaluation is carried out on each time period in the voice, the quality of the voice can be easily obtained, the accuracy of voice quality monitoring is improved, meanwhile, the working strength of voice testing is reduced, and the working efficiency is improved.

Compared with the prior art, the invention has the beneficial effects that:

1. in the invention, a section of 20 seconds of voice is acquired, then a section of voice with qualified quality evaluation is acquired and marked as standard voice, then the test voice is continuously played for one minute, a time point of each second is taken as a monitoring point, a voice sequence interval corresponding to the monitoring point in the test voice in one minute is acquired, the voice sequence interval is expressed as the interval duration of two bytes passing through the monitoring point when the bytes in the voice are played, and then a voice sequence interval corresponding to the monitoring point in the standard voice in one minute is acquired; comparing the voice sequence interval in the monitoring voice with the voice sequence interval in the standard voice, performing frame division on the monitoring voice, dividing the monitoring voice into a plurality of voice frames, acquiring the audio wavelength and frequency of the voice frames, then acquiring the parameter analysis coefficient Xi of the voice frames through a formula, constructing a parameter analysis coefficient set for the voice frames in the testing voice, and comparing the subset in the parameter analysis coefficient set with a parameter analysis coefficient threshold; the test voice is preprocessed, so that the qualification rate of the test voice is improved, the quality evaluation of unqualified voice is prevented, the working intensity is increased, and the working efficiency of voice quality evaluation is reduced;

2. in the invention, a quality evaluation model is constructed to carry out quality evaluation on voice data, a parameter analysis coefficient RS of a test voice and a quality analysis coefficient FXo corresponding to a quality qualified test voice are obtained, and then the quality evaluation model is constructed, namely the quality evaluation model formula is PJ ═ alpha (e) in the formula^RS×b1+e^FXo×b2)×t]^3.65Establishing a rectangular coordinate system by taking the independent variable time as an X axis and the quality evaluation coefficient as a Y axis, taking the integral point time as a point on the X axis, marking the inner point of the rectangular coordinate system corresponding to the integral point time on the X axis, and connecting the inner points in the rectangular coordinate system to construct a quality evaluation curve of the test voice; analyzing the quality evaluation curve, analyzing the corresponding integral point time in the total duration of the test voice, obtaining the slope of the quality evaluation curve between two adjacent integral point times, and analyzing the corresponding quality evaluation curveThe slopes will be compared; the quality evaluation model is built, the quality evaluation curve of the tested voice is built according to the model formula, quality evaluation is carried out on each time period in the voice, the quality of the voice can be easily obtained, the accuracy of voice quality monitoring is improved, meanwhile, the working strength of voice testing is reduced, and the working efficiency is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic block diagram of the present invention.

Detailed Description

The technical solutions of the present invention will be described clearly and completely with reference to the following embodiments, and it should be understood that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, a quality evaluation method based on voice data includes the following steps:

step one, voice preprocessing, namely preprocessing voice data so as to detect the voice data;

step two, data analysis, namely performing data analysis on the voice data which is preprocessed to obtain a quality analysis coefficient of the voice data;

thirdly, quality evaluation, namely constructing a quality evaluation model and carrying out quality evaluation on the voice data;

the voice data preprocessing in the first step comprises the following specific steps:

step S1: acquiring a section of 20-second voice, marking the voice as test voice, acquiring a section of voice with qualified quality evaluation, marking the voice as standard voice, continuously playing the test voice for one minute, acquiring a voice sequence interval corresponding to a monitoring point in the test voice within one minute by taking a time point of each second as the monitoring point, wherein the voice sequence interval is represented as the interval duration of two bytes passing through the monitoring point when the bytes in the voice are played, and then acquiring a voice sequence interval corresponding to the monitoring point in the standard voice within one minute;

step S2: comparing the voice sequence interval in the monitored voice with the voice sequence interval in the standard voice, if the voice sequence interval in the monitored voice is larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is unqualified, marking the tested voice sequence interval as abnormal voice of the sequence interval, if the voice sequence interval in the monitored voice is not larger than the voice sequence interval in the standard voice, judging that the tested voice sequence interval is qualified, and entering the step S3;

step S3: performing frame division on monitoring voice, dividing the monitoring voice into a plurality of voice frames, marking the voice frames as i, i is 1, 2, … …, n and n are positive integers, acquiring the audio wavelength and frequency of the voice frames, respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi, and then respectively marking the audio wavelength and frequency of the voice frames as Bi and Pi through a formula Xi is beta (Bi x a1+ Pi x a2)²Acquiring a parameter analysis coefficient Xi of the voice frame, wherein a1 and a2 are both proportional coefficients, and beta is an error correction factor and takes the value of 1.36;

step S4: constructing a parameter analysis coefficient set for a voice frame in the test voice, comparing subsets in the parameter analysis coefficient set with a parameter analysis coefficient threshold, if the subsets are more than or equal to the parameter analysis coefficient threshold, marking the corresponding subsets as qualified subsets, and if the subsets are less than the parameter analysis coefficient threshold, marking the corresponding subsets as unqualified subsets; if the number of qualified subsets in the parameter analysis coefficient set is larger than that of unqualified subsets, acquiring the average value of the parameter analysis coefficient set and marking the average value as a parameter analysis coefficient RS of the test voice, otherwise, marking the corresponding test voice as unqualified-parameter voice; the test voice is preprocessed, so that the qualification rate of the test voice is improved, the quality evaluation of unqualified voice is prevented, the working intensity is increased, and the working efficiency of voice quality evaluation is reduced;

the specific steps for obtaining the quality analysis coefficient of the voice data in the second step are as follows:

step SS 1: marking the test voice as o, o is 1, 2, … …, m is a positive integer, acquiring the audio signal-to-noise ratio of each monitoring point of the test voice, and marking the audio signal-to-noise ratio of each monitoring point of the test voice as XZo;

step SS 2: acquiring an audio cycle of each monitoring point of the test voice, and marking the audio cycle of each monitoring point of the test voice as ZQo;

step SS 3: acquiring the audio wavelength of each monitoring point of the test voice, and marking the audio wavelength of each monitoring point of the test voice as BCo;

step SS 4: by the formulaAcquiring a quality analysis coefficient FXo of the test voice, wherein s1, s2 and s3 are all proportionality coefficients, s1 is more than s2 is more than s3 is more than 0, and e is a natural constant;

step SS 5: the quality analysis coefficient FXo of the test speech is compared to a quality analysis coefficient threshold:

if the quality analysis coefficient FXo of the test voice is larger than or equal to the quality analysis coefficient threshold, judging that the quality of the test voice is qualified, generating a quality qualified signal and marking the corresponding test voice as a quality qualified test voice;

if the quality analysis coefficient FXo of the test voice is smaller than the quality analysis coefficient threshold value, judging that the quality of the test voice is unqualified, generating a signal with unqualified quality and marking the corresponding test voice as a test voice with unqualified quality;

and step three, constructing a quality evaluation model to evaluate the quality of the voice data, wherein the specifically constructing and evaluating steps are as follows:

step T1: obtaining a parameter analysis coefficient RS of the test voice and a quality analysis coefficient FXo corresponding to the quality qualified test voice, and then constructing a quality evaluation model, wherein the quality evaluation model formula is PJ ═ alpha (e)^RS×b1+e^FXo×b2)×t]^3.65Wherein, b1 and b2 are proportionality coefficients, b1 is larger than b2 is larger than 0, e is a natural constant, t is independent variable time, and PJ is a quality evaluation coefficient;

step T2: establishing a rectangular coordinate system by taking the independent variable time as an X axis and the quality evaluation coefficient as a Y axis, taking the integral point moment as a point on the X axis, marking the inner point of the rectangular coordinate system corresponding to the integral point moment on the X axis, and connecting the inner points in the rectangular coordinate system to construct a quality evaluation curve of the tested voice;

step T3: analyzing the quality evaluation curve, analyzing the corresponding integral point time in the total duration of the tested voice, obtaining the slope of the quality evaluation curve between two adjacent integral point times, and comparing the corresponding slope of the quality evaluation curve:

if the slope of the corresponding quality evaluation curve is less than 40 degrees, judging that the test voice corresponding to the time of the adjacent integral points is normal; if the slope of the corresponding quality evaluation curve is less than 65 degrees and less than 40 degrees, judging that the testing voice corresponding to the adjacent integral point time is abnormal, marking the testing voice corresponding to the adjacent integral point time as abnormal voice, generating a voice rectification signal and sending the voice rectification signal to a mobile phone terminal of a manager; quality evaluation is carried out on each time period in the voice, the quality of the voice can be easily obtained, the accuracy of voice quality monitoring is improved, meanwhile, the working strength of voice testing is reduced, and the working efficiency is improved.

The working principle of the invention is as follows:

a quality evaluation method based on voice data comprises the following steps: voice preprocessing, namely preprocessing voice data so as to detect the voice data; data analysis, namely performing data analysis on the preprocessed voice data to obtain a quality analysis coefficient of the voice data; quality evaluation, namely constructing a quality evaluation model and carrying out quality evaluation on voice data;

acquiring a section of 20-second voice, marking the voice as test voice, acquiring a section of voice with qualified quality evaluation, marking the voice as standard voice, continuously playing the test voice for one minute, acquiring a voice sequence interval corresponding to a monitoring point in the test voice within one minute by taking a time point of each second as the monitoring point, wherein the voice sequence interval is represented as the interval duration of two bytes passing through the monitoring point when the bytes in the voice are played, and then acquiring a voice sequence interval corresponding to the monitoring point in the standard voice within one minute; comparing the speech sequence interval in the monitored speech with the speech sequence interval in the standard speech, performing frame division on the monitored speech, dividing the monitored speech into a plurality of speech frames, marking the speech frames as i, i is 1, 2, … …, n, n is a positive integer, obtaining the audio wavelength and frequency of the speech frames, then obtaining the parameter analysis coefficients X i of the speech frames through a formula, constructing a parameter analysis coefficient set for the speech frames in the tested speech, and comparing the subset in the parameter analysis coefficient set with a parameter analysis coefficient threshold.

The above formulas are all calculated by taking the numerical value of the dimension, the formula is a formula which obtains the latest real situation by acquiring a large amount of data and performing software simulation, and the preset parameters in the formula are set by the technical personnel in the field according to the actual situation.

The foregoing is merely exemplary and illustrative of the present invention and various modifications, additions and substitutions may be made by those skilled in the art to the specific embodiments described without departing from the scope of the invention as defined in the following claims.