Voice evaluation method, device and electronic equipment

Abstract

The present disclosure provides a speech evaluation method, device and electronic device, the method comprising: performing speech segmentation on a speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated; determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment, the at least one evaluation index value being used to indicate the energy proportion of clean speech in the speech to be evaluated; determining the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value. The embodiments of the present disclosure can improve the efficiency of speech evaluation and can improve the accuracy of speech evaluation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a speech evaluation method, device and electronic equipment. Background Art

[0002] In the process of model training, a large amount of clean data is required as training samples. Clean data refers to speech data that is theoretically noise-free. The less noise in the speech data, the cleaner the speech is, and the more suitable it is as clean data for model training. Currently, clean data for model training is obtained by recording speech in a quiet environment. In order to avoid low cleanliness of recorded speech, the recorded speech is usually played and the cleanliness of the speech is judged by human ears, which has low efficiency in speech evaluation. Summary of the invention

[0003] The embodiments of the present disclosure provide a speech evaluation method, device, and electronic device to solve the problem of low efficiency of speech evaluation caused by human ears judging the cleanliness of speech in the prior art.

[0004] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0005] In a first aspect, an embodiment of the present disclosure provides a speech evaluation method, the method comprising:

[0006] Performing speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated;

[0007] Determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate the energy proportion of the clean speech in the speech to be evaluated in the speech to be evaluated;

[0008] The speech cleanliness of the speech to be evaluated is determined based on the at least one evaluation index value.

[0009] In a second aspect, an embodiment of the present disclosure provides a speech evaluation device, the device comprising:

[0010] A segmentation module, used to perform speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated;

[0011] A first determination module, configured to determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate an energy proportion of clean speech in the speech to be evaluated in the speech to be evaluated;

[0012] The second determination module is used to determine the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value.

[0013] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps in the speech evaluation method described in the first aspect.

[0014] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the speech evaluation method described in the first aspect are implemented.

[0015] In the disclosed embodiment, the speech to be evaluated is segmented to obtain the first speech segment and the first non-speech segment corresponding to the speech to be evaluated; at least one evaluation index value is determined based on the first speech segment and / or the first non-speech segment, and the at least one evaluation index value is used to indicate the energy proportion of the clean speech in the speech to be evaluated; the speech cleanliness of the speech to be evaluated is determined based on the at least one evaluation index value. In this way, by determining the speech cleanliness of the speech to be evaluated through the at least one evaluation index value, automatic evaluation of speech cleanliness can be achieved, thereby improving the efficiency of speech evaluation and the accuracy of speech evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the description of the embodiments of the present disclosure will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 is a flow chart of a speech evaluation method provided by an embodiment of the present disclosure;

[0018] Figure 2 is a structural schematic diagram of a speech evaluation device provided by an embodiment of the present disclosure;

[0019] Figure 3 It is a structural schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] In the embodiments of the present disclosure, electronic devices include but are not limited to mobile phones, tablet computers, laptop computers, PDAs, vehicle-mounted mobile terminals, wearable devices, and pedometers.

[0022] See also Figure 1 , Figure 1 is a flow chart of a speech evaluation method provided by an embodiment of the present disclosure, such as Figure 1 As shown, the following steps are included:

[0023] Step 101: Segment the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated.

[0024] The first speech segment corresponding to the speech to be evaluated may include at least one speech segment obtained by speech segmentation of the speech to be evaluated. The first non-speech segment corresponding to the speech to be evaluated may include at least one non-speech segment obtained by speech segmentation of the speech to be evaluated.

[0025] In one implementation, the first speech segment corresponding to the speech to be evaluated may include all speech segments obtained by speech segmentation of the speech to be evaluated. The first non-speech segment corresponding to the speech to be evaluated may include all non-speech segments obtained by speech segmentation of the speech to be evaluated.

[0026] In one implementation, the frame length and frame shift of the speech to be evaluated can be determined, and the speech to be evaluated can be framed. After framing, a voice activity detection (VAD) algorithm can be used to perform speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated.

[0027] Step 102: determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate the energy proportion of clean speech in the speech to be evaluated.

[0028] Among them, the at least one evaluation index value may include one or more of a first evaluation index value, a second evaluation index value, a third evaluation index value and a fourth evaluation index value. The first evaluation index value is determined based on a second speech segment, and the second speech segment is obtained after low-pass filtering based on the first speech segment. The second evaluation index value is determined based on a second non-speech segment, and the second non-speech segment is obtained after low-pass filtering based on the first non-speech segment. The third evaluation index value is determined based on the data points corresponding to the fundamental frequency of the first speech segment and the multiples of the fundamental frequency. The fourth evaluation index value is determined based on the energy value of the first speech segment and the energy value of the first non-speech segment.

[0029] Step 103: Determine the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value.

[0030] Among them, the larger the value of the at least one evaluation index is, the higher the speech cleanliness of the speech to be evaluated is.

[0031] It should be noted that the current model training method for speech usually uses the supervised training method, and the selection of clean data in supervised training plays a very important role in the effect of the model and can directly affect the effect of the model. The smaller the noise of the speech data, the cleaner the speech is, and the more suitable it is as clean data for model training.

[0032] In one embodiment, the at least one evaluation index value may include a first evaluation index value, a second evaluation index value, a third evaluation index value, and a fourth evaluation index value. The data to be evaluated that satisfies the first evaluation index value greater than a first preset threshold, the second evaluation index value greater than a second preset threshold, the third evaluation index value greater than a third preset threshold, and the fourth evaluation index value greater than a fourth preset threshold can be used for model training, which can improve the effect of model training. The first preset threshold, the second preset threshold, the third preset threshold, and the fourth preset threshold can be set according to actual needs.

[0033] In the disclosed embodiment, the speech to be evaluated is segmented to obtain the first speech segment and the first non-speech segment corresponding to the speech to be evaluated; at least one evaluation index value is determined based on the first speech segment and / or the first non-speech segment, and the at least one evaluation index value is used to indicate the energy proportion of the clean speech in the speech to be evaluated; the speech cleanliness of the speech to be evaluated is determined based on the at least one evaluation index value. In this way, by determining the speech cleanliness of the speech to be evaluated through the at least one evaluation index value, automatic evaluation of speech cleanliness can be achieved, thereby improving the efficiency of speech evaluation and the accuracy of speech evaluation.

[0034] Optionally, the at least one evaluation index value includes at least one of the following:

[0035] The first evaluation indicator value;

[0036] The second evaluation indicator value;

[0037] The third evaluation index value;

[0038] The fourth evaluation index value;

[0039] Wherein, the first evaluation index value is determined based on a second speech segment, and the second speech segment is obtained after low-pass filtering based on the first speech segment;

[0040] The second evaluation index value is determined based on a second non-speech segment, and the second non-speech segment is obtained after low-pass filtering based on the first non-speech segment;

[0041] The third evaluation index value is determined based on the data points corresponding to the fundamental frequency of the first speech segment and the multiple frequencies of the fundamental frequency;

[0042] The fourth evaluation index value is determined based on an energy value of the first speech segment and an energy value of the first non-speech segment.

[0043] The first evaluation index value may be a ratio of the energy mean of the second speech segment to the energy mean of the speech to be evaluated; or, the first evaluation index value may be a ratio of the energy mean of the second speech segment to the energy mean of a third speech segment, where the third speech segment is obtained after high-pass filtering based on the first speech segment; or, the first evaluation index value may be the energy mean of the second speech segment; and so on. This embodiment does not limit this.

[0044] In addition, the second evaluation index value may be a ratio of the energy mean of the second non-speech segment to the energy mean of the speech to be evaluated; or, the second evaluation index value may be a ratio of the energy mean of the second non-speech segment to the energy mean of the third non-speech segment, the third non-speech segment being obtained after high-pass filtering based on the first non-speech segment; or, the second evaluation index value may be the energy mean of the second non-speech segment; and so on. This embodiment does not limit this.

[0045] In addition, the third evaluation index value may be a ratio of the energy mean of the first data point corresponding to the fundamental frequency of the first speech segment and the multiple of the fundamental frequency to the energy mean of the speech to be evaluated; or, the third evaluation index value may be a ratio of the energy mean of the first data point to the energy mean of the second data point, the second data point being the second data point in the first speech segment other than the first data point; or, the third evaluation index value may be the energy mean of the first data point; and so on. This embodiment is not limited to this.

[0046] In addition, the fourth evaluation index value may be a ratio of the energy value of the first speech segment to the energy value of the first non-speech segment; or the fourth evaluation index value may be a ratio of the energy mean of the first speech segment to the energy mean of the first non-speech segment; etc. This embodiment does not limit this.

[0047] In this implementation, by using at least one of the first evaluation index value, the second evaluation index value, the third evaluation index value and the fourth evaluation index value, the speech cleanliness of the speech to be evaluated can be evaluated from one or more dimensions, thereby improving the accuracy of speech evaluation.

[0048] Optionally, determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment includes:

[0049] Performing low-pass filtering on the first speech segment to obtain a second speech segment;

[0050] Performing high-pass filtering on the first speech segment to obtain a third speech segment;

[0051] Determine a first ratio, wherein the first ratio is a ratio of an energy mean of the second speech segment to an energy mean of the third speech segment;

[0052] The at least one evaluation index value includes a first evaluation index value, and the first evaluation index value is the first ratio.

[0053] In addition, the second speech segment may include multiple third data points. The energy mean of the second speech segment may be the mean of the squares of the amplitudes of the multiple third data points; or the energy mean of the second speech segment may be obtained as follows: determine the difference between the amplitude of each third data point and the first amplitude mean, calculate the square value of the difference corresponding to each third data point, and use the mean of the square values ​​corresponding to the multiple third data points as the energy mean of the second speech segment, and the first amplitude mean is the mean of the amplitudes of the multiple third data points. For example, the energy mean Lspeech of the second speech segment may be obtained by the following calculation formula:

[0054]

[0055] Among them, X i is the difference between the amplitude of the ith third data point and the mean of the first amplitudes, and m is the number of third data points included in the second speech segment.

[0056] In addition, the third speech segment may include multiple fourth data points. The energy mean of the third speech segment may be the mean of the squares of the amplitudes of the multiple fourth data points; or the energy mean of the third speech segment may be obtained as follows: determine the difference between the amplitude of each fourth data point and the second amplitude mean, calculate the square value of the difference corresponding to each fourth data point, and use the mean of the square values ​​corresponding to the multiple fourth data points as the energy mean of the third speech segment, and the second amplitude mean is the mean of the amplitudes of the multiple fourth data points. For example, the energy mean Hspeech of the third speech segment may be obtained by the following calculation formula:

[0057]

[0058] Among them, X i is the difference between the amplitude of the ith fourth data point and the mean of the second amplitudes, and m is the number of fourth data points included in the third speech segment.

[0059] It should be noted that this embodiment does not limit the filter cutoff frequency of high-pass filtering and low-pass filtering. For example, the filter cutoff frequency may be 3KHZ, and high-pass filtering filters out signals below 3KHZ, and low-pass filtering filters out signals above 3KHZ.

[0060] In this implementation, the first speech segment is subjected to low-pass filtering to obtain a second speech segment; the first speech segment is subjected to high-pass filtering to obtain a third speech segment; a first ratio is determined, the first ratio being the ratio of the energy mean of the second speech segment to the energy mean of the third speech segment; wherein the at least one evaluation index value includes a first evaluation index value, the first evaluation index value being the first ratio. In this way, since low-pass filtering can filter out high-frequency noise, and high-pass filtering can filter out low-frequency signals and retain high-frequency noise, the first ratio can reflect the speech cleanliness of the speech to be evaluated. When the first ratio is larger, the speech cleanliness of the speech to be evaluated is higher. From the dimension of the speech segment of the speech to be evaluated, determining the speech cleanliness of the speech to be evaluated by the first ratio can improve the efficiency of speech evaluation and can improve the accuracy of speech evaluation.

[0061] Optionally, determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment includes:

[0062] Performing low-pass filtering on the first non-speech segment to obtain a second non-speech segment;

[0063] Performing high-pass filtering on the first non-speech segment to obtain a third non-speech segment;

[0064] Determine a second ratio, wherein the second ratio is a ratio of an energy mean of the second non-speech segment to an energy mean of the third non-speech segment;

[0065] The at least one evaluation index value includes a second evaluation index value, and the second evaluation index value is the second ratio.

[0066] In addition, the second non-speech segment may include multiple fifth data points. The energy mean of the second non-speech segment may be the mean of the squares of the amplitudes of the multiple fifth data points; or the energy mean of the second non-speech segment may be obtained as follows: determine the difference between the amplitude of each fifth data point and the third amplitude mean, calculate the square value of the difference corresponding to each fifth data point, and use the mean of the square values ​​corresponding to the multiple fifth data points as the energy mean of the second non-speech segment, and the third amplitude mean is the mean of the amplitudes of the multiple fifth data points. For example, the energy mean Lnoise of the second non-speech segment may be obtained by the following calculation formula:

[0067]

[0068] Among them, X i is the difference between the amplitude of the ith fifth data point and the third amplitude mean, and m is the number of fifth data points included in the second non-speech segment.

[0069] In addition, the third non-speech segment may include multiple sixth data points. The energy mean of the third non-speech segment may be the mean of the squares of the amplitudes of the multiple sixth data points; or the energy mean of the third non-speech segment may be obtained as follows: determine the difference between the amplitude of each sixth data point and the fourth amplitude mean, calculate the square value of the difference corresponding to each sixth data point, and use the mean of the square values ​​corresponding to the multiple sixth data points as the energy mean of the third non-speech segment, and the fourth amplitude mean is the mean of the amplitudes of the multiple sixth data points. For example, the energy mean Hnoise of the third non-speech segment may be obtained by the following calculation formula:

[0070]

[0071] Among them, X i is the difference between the amplitude of the ith sixth data point and the fourth amplitude mean, and m is the number of sixth data points included in the third non-speech segment.

[0072] In this implementation, the first non-speech segment is subjected to low-pass filtering to obtain a second non-speech segment; the first non-speech segment is subjected to high-pass filtering to obtain a third non-speech segment; a second ratio is determined, the second ratio being the ratio of the energy mean of the second non-speech segment to the energy mean of the third non-speech segment; wherein the at least one evaluation index value includes a second evaluation index value, the second evaluation index value being the second ratio. In this way, since the low-pass filtering process can filter out high-frequency noise, and the high-pass filtering process can filter out low-frequency signals and retain high-frequency noise, the second ratio can reflect the speech cleanliness of the speech to be evaluated. When the second ratio is larger, the speech cleanliness of the speech to be evaluated is higher. From the dimension of the non-speech segment of the speech to be evaluated, determining the speech cleanliness of the speech to be evaluated through the second ratio can improve the efficiency of speech evaluation and can improve the accuracy of speech evaluation.

[0073] Optionally, determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment includes:

[0074] Determine a first data point in the first speech segment, where the first data point is a data point corresponding to a fundamental frequency of the first speech segment and a multiple of the fundamental frequency;

[0075] Determine a second data point, where the second data point is a data point in the first speech segment other than the first data point;

[0076] Determine a third ratio, the third ratio being a ratio of an energy mean of the first data point to an energy mean of the second data point;

[0077] The at least one evaluation index value includes a third evaluation index value, and the third evaluation index value is the third ratio.

[0078] Taking the fundamental frequency of the first speech segment as 65Hz as an example, the multiples of the fundamental frequency are 2*65Hz, 3*65Hz, 4*65Hz...; and so on. The first data point is the data point where the frequencies of 65Hz, 2*65Hz, 3*65Hz, 4*65Hz,... are located in the first speech segment.

[0079] It should be noted that the method for determining the first data point can be as follows: the spectrum of the first speech segment can be obtained, and the clef of the spectrum is: 0, 1, 2, 3, 4, ..., 256. The frequencies corresponding to the clef of the spectrum are: 0*32.5, 1*32.5, ..., 256*32.5Hz. Assuming that the fundamental frequency is 65Hz, the multiples are: 2*65, 3*65, ..., etc., then the corresponding clefs can be obtained as: 65 / 32.5=2, 2*65 / 32.5=4, 3*65 / 32.5=6, ..., etc. The points at the clefs are taken as the first data points.

[0080] In addition, the number of data points of the first data point can be multiple. The energy mean of the first data point can be the mean of the squares of the amplitudes of the first data points in the first speech segment; or the energy mean of the first data point can be obtained as follows: determine the difference between the amplitude of each first data point and the fifth amplitude mean, calculate the square value of the difference corresponding to each first data point, and use the mean of the square values ​​corresponding to the first data points in the first speech segment as the energy mean of the first data point, and the fifth amplitude mean is the mean of the amplitudes of the first data points in the first speech segment. For example, the energy mean Eb of the first data point can be obtained by the following calculation formula:

[0081]

[0082] Among them, X i is the difference between the amplitude of the ith first data point and the fifth amplitude mean, and m is the number of first data points included in the first speech segment.

[0083] In addition, the number of data points of the second data point can be multiple. The energy mean of the second data point can be the mean of the squares of the amplitudes of the second data points in the first speech segment; or the energy mean of the second data point can be obtained as follows: determine the difference between the amplitude of each second data point and the sixth amplitude mean, calculate the square value of the difference corresponding to each second data point, and use the mean of the square values ​​corresponding to the second data points in the first speech segment as the energy mean of the second data point, and the sixth amplitude mean is the mean of the amplitudes of the second data points in the first speech segment. For example, the energy mean Eo of the second data point can be obtained by the following calculation formula:

[0084]

[0085] Among them, X i is the difference between the amplitude of the ith second data point and the sixth amplitude mean, and m is the number of second data points included in the first speech segment.

[0086] In this implementation, the first data point in the first speech segment is determined, the first data point is the data point corresponding to the fundamental frequency of the first speech segment and the multiple frequency of the fundamental frequency; the second data point is determined, the second data point is the data point in the first speech segment other than the first data point; the third ratio is determined, the third ratio is the ratio of the energy mean of the first data point to the energy mean of the second data point; wherein the at least one evaluation index value includes a third evaluation index value, the third evaluation index value is the third ratio. In this way, since the first data point corresponding to the fundamental frequency and the multiple frequency of the fundamental frequency is likely to be clean data, and the second data point is likely to be noise data, the third ratio can reflect the speech cleanliness of the speech to be evaluated. When the third ratio is larger, the speech cleanliness of the speech to be evaluated is higher. From the frequency dimension of the speech to be evaluated, the speech cleanliness of the speech to be evaluated is determined by the third ratio, which can improve the efficiency of speech evaluation and the accuracy of speech evaluation.

[0087] Optionally, determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment includes:

[0088] Respectively obtaining an energy mean of the first speech segment and an energy mean of the first non-speech segment;

[0089] Determine a fourth ratio, wherein the fourth ratio is a ratio of an energy mean of the first speech segment to an energy mean of the first non-speech segment;

[0090] The at least one evaluation index value includes a fourth evaluation index value, and the fourth evaluation index value is the fourth ratio.

[0091] In addition, the first speech segment may include multiple seventh data points. The energy mean of the first speech segment may be the average of the squares of the amplitudes of multiple seventh data points; or the energy mean of the first speech segment may be obtained as follows: determine the difference between the amplitude of each seventh data point and the seventh amplitude mean, calculate the square value of the difference corresponding to each seventh data point, and use the average of the square values ​​corresponding to multiple seventh data points as the energy mean of the first speech segment, and the seventh amplitude mean is the average of the amplitudes of the multiple seventh data points. For example, the energy mean Espeech of the first speech segment may be obtained by the following calculation formula:

[0092]

[0093] Among them, X i is the difference between the amplitude of the ith seventh data point and the mean of the seventh amplitudes, and m is the number of seventh data points included in the first speech segment.

[0094] In addition, the first non-speech segment may include multiple eighth data points. The energy mean of the first non-speech segment may be the mean of the squares of the amplitudes of multiple eighth data points; or the energy mean of the first non-speech segment may be obtained as follows: determine the difference between the amplitude of each eighth data point and the mean of the eighth amplitude, calculate the square value of the difference corresponding to each eighth data point, and use the mean of the square values ​​corresponding to multiple eighth data points as the energy mean of the first non-speech segment, and the eighth amplitude mean is the mean of the amplitudes of the multiple eighth data points. For example, the energy mean Enoise of the first non-speech segment may be obtained by the following calculation formula:

[0095]

[0096] Among them, X i is the difference between the amplitude of the ith eighth data point and the eighth amplitude mean, and m is the number of eighth data points included in the first non-speech segment.

[0097] In this implementation, the energy mean of the first speech segment and the energy mean of the first non-speech segment are obtained respectively; a fourth ratio is determined, and the fourth ratio is the ratio of the energy mean of the first speech segment to the energy mean of the first non-speech segment; wherein the at least one evaluation index value includes a fourth evaluation index value, and the fourth evaluation index value is the fourth ratio. In this way, the speech cleanliness of the speech to be evaluated can be reflected by the fourth ratio. When the fourth ratio is larger, the speech cleanliness of the speech to be evaluated is higher, which can improve the efficiency of speech evaluation and the accuracy of speech evaluation.

[0098] Optionally, determining the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value includes:

[0099] Obtaining a first weight value, a second weight value, a third weight value, and a fourth weight value;

[0100] Determine a target evaluation index value based on the product of the first evaluation index value and a first weight, the product of the second evaluation index value and a second weight value, the product of the third evaluation index value and a third weight value, and the product of the fourth evaluation index value and a fourth weight value;

[0101] The speech cleanliness of the speech to be evaluated is determined based on the target evaluation index value.

[0102] Among them, the first weight value, the second weight value, the third weight value and the fourth weight value can all be preset. For example, the first weight value can be set to 0.3, the second weight value can be set to 0.2, the third weight value can be set to 0.3, and the fourth weight value can be set to 0.2; or, the first weight value can be set to 5, the second weight value can be set to 4, the third weight value can be set to 2, and the fourth weight value can be set to 2; or, the first weight value can be set to 30, the second weight value can be set to 28, the third weight value can be set to 21, and the fourth weight value can be set to 10; etc., this embodiment does not limit this.

[0103] In one implementation, the sum of the first weight value, the second weight value, the third weight value, and the fourth weight value may be 1.

[0104] In one embodiment, the target evaluation index value Q = a*Js+b*Jn+c*Jf+d*Je, wherein Js is the first ratio, Jn is the second ratio, Jf is the third ratio, Je is the fourth ratio, a is the first weight value, b is the second weight value, c is the third weight value, and d is the fourth weight value.

[0105] In addition, the larger the target evaluation index value is, the higher the speech cleanliness of the speech to be evaluated is. The data to be evaluated whose target evaluation index value is greater than the preset threshold can be used for model training, which can improve the effect of model training. The preset threshold can be set according to actual needs.

[0106] In this implementation, the target evaluation index value is determined based on the product of the first evaluation index value and the first weight, the product of the second evaluation index value and the second weight value, the product of the third evaluation index value and the third weight value, and the product of the fourth evaluation index value and the fourth weight value; the speech cleanliness of the speech to be evaluated is determined based on the target evaluation index value. In this way, the speech cleanliness of the speech to be evaluated is comprehensively evaluated from multiple dimensions, so that the speech cleanliness of the speech to be evaluated can be evaluated more comprehensively and accurately.

[0107] See also Figure 2 , Figure 2 is a structural diagram of a speech evaluation device provided by an embodiment of the present disclosure, such as Figure 2 As shown, the speech evaluation device 200 includes:

[0108] The segmentation module 201 is used to perform speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated;

[0109] A first determination module 202, configured to determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate an energy proportion of clean speech in the speech to be evaluated in the speech to be evaluated;

[0110] The second determination module 203 is configured to determine the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value.

[0111] Optionally, the at least one evaluation index value includes at least one of the following:

[0112] The first evaluation indicator value;

[0113] The second evaluation indicator value;

[0114] The third evaluation index value;

[0115] The fourth evaluation index value;

[0116] Wherein, the first evaluation index value is determined based on a second speech segment, and the second speech segment is obtained after low-pass filtering based on the first speech segment;

[0117] The second evaluation index value is determined based on a second non-speech segment, and the second non-speech segment is obtained after low-pass filtering based on the first non-speech segment;

[0118] The third evaluation index value is determined based on the data points corresponding to the fundamental frequency of the first speech segment and the multiple frequencies of the fundamental frequency;

[0119] The fourth evaluation index value is determined based on an energy value of the first speech segment and an energy value of the first non-speech segment.

[0120] Optionally, the first determining module is specifically configured to:

[0121] Performing low-pass filtering on the first speech segment to obtain a second speech segment;

[0122] Performing high-pass filtering on the first speech segment to obtain a third speech segment;

[0123] Determine a first ratio, wherein the first ratio is a ratio of an energy mean of the second speech segment to an energy mean of the third speech segment;

[0124] The at least one evaluation index value includes a first evaluation index value, and the first evaluation index value is the first ratio.

[0125] Optionally, the first determining module is specifically configured to:

[0126] Performing low-pass filtering on the first non-speech segment to obtain a second non-speech segment;

[0127] Performing high-pass filtering on the first non-speech segment to obtain a third non-speech segment;

[0128] Determine a second ratio, wherein the second ratio is a ratio of an energy mean of the second non-speech segment to an energy mean of the third non-speech segment;

[0129] The at least one evaluation index value includes a second evaluation index value, and the second evaluation index value is the second ratio.

[0130] Optionally, the first determining module is specifically configured to:

[0131] Determine a first data point in the first speech segment, where the first data point is a data point corresponding to a fundamental frequency of the first speech segment and a multiple of the fundamental frequency;

[0132] Determine a second data point, where the second data point is a data point in the first speech segment other than the first data point;

[0133] Determine a third ratio, the third ratio being a ratio of an energy mean of the first data point to an energy mean of the second data point;

[0134] The at least one evaluation index value includes a third evaluation index value, and the third evaluation index value is the third ratio.

[0135] Optionally, the first determining module is specifically configured to:

[0136] Respectively obtaining an energy mean of the first speech segment and an energy mean of the first non-speech segment;

[0137] Determine a fourth ratio, wherein the fourth ratio is a ratio of an energy mean of the first speech segment to an energy mean of the first non-speech segment;

[0138] The at least one evaluation index value includes a fourth evaluation index value, and the fourth evaluation index value is the fourth ratio.

[0139] Optionally, the first determining module is specifically configured to:

[0140] Obtaining a first weight value, a second weight value, a third weight value, and a fourth weight value;

[0141] Determine a target evaluation index value based on the product of the first evaluation index value and a first weight, the product of the second evaluation index value and a second weight value, the product of the third evaluation index value and a third weight value, and the product of the fourth evaluation index value and a fourth weight value;

[0142] The speech cleanliness of the speech to be evaluated is determined based on the target evaluation index value.

[0143] The speech evaluation device enables Figure 1 The various processes implemented in the method embodiment can achieve the same technical effect, and to avoid repetition, they will not be described again here.

[0144] like Figure 3 As shown, the embodiment of the present disclosure further provides an electronic device 300, including: a memory 302, a processor 301, and a program stored in the memory 302 and executable on the processor 301. When the program is executed by the processor 301, each process of the above-mentioned speech evaluation method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0145] The embodiment of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above-mentioned speech evaluation method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0146] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0147] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in the various embodiments of the present disclosure.

[0148] The embodiments of the present disclosure are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A speech evaluation method, characterized in that: The method comprises: Performing speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated; Determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate the energy proportion of the clean speech in the speech to be evaluated in the speech to be evaluated; Determining the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value; The determining at least one evaluation index value based on the first speech segment and / or the first non-speech segment includes: Determine a first data point in the first speech segment, where the first data point is a data point corresponding to a fundamental frequency of the first speech segment and a multiple of the fundamental frequency; Determine a second data point, where the second data point is a data point in the first speech segment other than the first data point; Determine a third ratio, the third ratio being a ratio of an energy mean of the first data point to an energy mean of the second data point; The at least one evaluation index value includes a third evaluation index value, and the third evaluation index value is the third ratio.

2. The method according to claim 1, characterized in that The at least one evaluation index value also includes at least one of the following: The first evaluation indicator value; The second evaluation indicator value; The fourth evaluation index value; Wherein, the first evaluation index value is determined based on a second speech segment, and the second speech segment is obtained after low-pass filtering based on the first speech segment; The second evaluation index value is determined based on a second non-speech segment, and the second non-speech segment is obtained after low-pass filtering based on the first non-speech segment; The fourth evaluation index value is determined based on an energy value of the first speech segment and an energy value of the first non-speech segment.

3. The method according to claim 1 or 2, characterized in that: The determining of at least one evaluation index value based on the first speech segment and / or the first non-speech segment further includes: Performing low-pass filtering on the first speech segment to obtain a second speech segment; Performing high-pass filtering on the first speech segment to obtain a third speech segment; Determine a first ratio, wherein the first ratio is a ratio of an energy mean of the second speech segment to an energy mean of the third speech segment; The at least one evaluation index value further includes a first evaluation index value, and the first evaluation index value is the first ratio.

4. The method according to claim 1 or 2, characterized in that: The determining of at least one evaluation index value based on the first speech segment and / or the first non-speech segment further includes: Performing low-pass filtering on the first non-speech segment to obtain a second non-speech segment; Performing high-pass filtering on the first non-speech segment to obtain a third non-speech segment; Determine a second ratio, wherein the second ratio is a ratio of an energy mean of the second non-speech segment to an energy mean of the third non-speech segment; The at least one evaluation index value further includes a second evaluation index value, and the second evaluation index value is the second ratio.

5. The method according to claim 1 or 2, characterized in that: The determining of at least one evaluation index value based on the first speech segment and / or the first non-speech segment further includes: Respectively obtaining an energy mean of the first speech segment and an energy mean of the first non-speech segment; Determine a fourth ratio, wherein the fourth ratio is a ratio of an energy mean of the first speech segment to an energy mean of the first non-speech segment; The at least one evaluation index value further includes a fourth evaluation index value, and the fourth evaluation index value is the fourth ratio.

6. The method according to claim 2, characterized in that The determining the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value includes: Obtaining a first weight value, a second weight value, a third weight value, and a fourth weight value; Determine a target evaluation index value based on the product of the first evaluation index value and a first weight, the product of the second evaluation index value and a second weight value, the product of the third evaluation index value and a third weight value, and the product of the fourth evaluation index value and a fourth weight value; The speech cleanliness of the speech to be evaluated is determined based on the target evaluation index value.

7. A speech evaluation device, characterized in that: The device comprises: A segmentation module, used to perform speech segmentation on the speech to be evaluated to obtain a first speech segment and a first non-speech segment corresponding to the speech to be evaluated; A first determination module, configured to determine at least one evaluation index value based on the first speech segment and / or the first non-speech segment, wherein the at least one evaluation index value is used to indicate an energy proportion of clean speech in the speech to be evaluated in the speech to be evaluated; A second determination module, configured to determine the speech cleanliness of the speech to be evaluated based on the at least one evaluation index value; The first determining module is specifically used for: Determine a first data point in the first speech segment, where the first data point is a data point corresponding to a fundamental frequency of the first speech segment and a multiple of the fundamental frequency; Determine a second data point, where the second data point is a data point in the first speech segment other than the first data point; Determine a third ratio, the third ratio being a ratio of an energy mean of the first data point to an energy mean of the second data point; The at least one evaluation index value includes a third evaluation index value, and the third evaluation index value is the third ratio.

8. An electronic device, characterized in that: include: A memory, a processor, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps in the speech evaluation method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the speech evaluation method according to any one of claims 1 to 6 are implemented.

Images