Audio processing method and device, equipment, storage medium and program product

Abstract

The invention discloses an audio processing method and device, equipment, a storage medium and a program product, and relates to the technical field of data processing, and the disclosed audio processing method comprises the steps: obtaining a plurality of to-be-processed audio files; performing quality evaluation on each to-be-processed audio file to obtain a quality evaluation result of each to-be-processed audio file; under the condition that the quality evaluation result is effective audio, taking the corresponding audio file to be processed as an effective audio file; when the quality evaluation result is invalid audio, deleting the corresponding audio file to be processed; and under the condition that the quality evaluation result is other evaluation results except the valid audio and the invalid audio, carrying out self-optimization processing on the corresponding audio file to be processed. According to the method, due to the fact that self-optimization processing is carried out on other data except the valid data and the invalid data, the quality of the whole data set is improved, and the problem of how to improve the overall quality of the multi-modal annotation data of the large model is solved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an audio processing method, device, equipment, storage medium and program product. Background Art

[0002] In the related art, the quality of multimodal annotation data of large models is evaluated and the erroneous and invalid data in the data set are deleted to achieve quality monitoring of multimodal annotation data. However, simply deleting erroneous and invalid data still cannot guarantee the overall quality of the data. Therefore, the problem of how to improve the overall quality of multimodal annotation data of large models needs to be solved urgently. Summary of the invention

[0003] The main purpose of this application is to provide an audio processing method, device, equipment, storage medium and program product, aiming to solve the technical problem of how to improve the overall quality of multimodal annotation data of large models.

[0004] To achieve the above purpose, the present application proposes an audio processing method, which includes:

[0005] Get multiple audio files to be processed;

[0006] Performing quality assessment on each audio file to be processed to obtain a quality assessment result of each audio file to be processed;

[0007] When the quality assessment result is valid audio, the corresponding audio file to be processed is regarded as a valid audio file;

[0008] If the quality assessment result is invalid audio, the corresponding audio file to be processed will be deleted;

[0009] When the quality assessment result is other than valid audio and invalid audio, the corresponding audio file to be processed is self-optimized.

[0010] In one embodiment, the plurality of audio files to be processed include an audio file that is too short, the audio duration of which is less than a preset duration and / or the length of the annotated text is less than a preset text length, and the quality assessment result includes a repairability score of the audio file that is too short;

[0011] When the quality assessment result is other than valid audio and invalid audio, the step of performing self-optimization processing on the corresponding audio file to be processed includes:

[0012] splicing the target annotated text of the target too short audio file whose repairability score is greater than or equal to the preset repairability score with the selected annotated text of the selected audio file to obtain a first spliced ​​text, wherein, when the repairability score is greater than or equal to the preset repairability score, the quality assessment result is other assessment results except for valid audio and invalid audio, and the selected audio file includes at least one to-be-processed audio file whose subscript number is adjacent to the target too short audio file in the audio group where the target too short audio file is located;

[0013] According to the punctuation positions in the first concatenated text, determining the target concatenated text where the target annotated text is located from the first concatenated text;

[0014] Determining a first selected spliced ​​text and a second selected spliced ​​text according to the position of the target annotated text in the target spliced ​​text and the target spliced ​​text;

[0015] Determine a target audio file from the selected audio files according to a first integrity probability of the first selected concatenated text and a second integrity probability of the second selected concatenated text;

[0016] The target audio file that is too short is merged with the target audio file to obtain a merged audio file, so as to perform self-optimization processing on the target audio file that is too short.

[0017] In one embodiment, after the step of merging the target short audio file with the target audio file to obtain a merged audio file, the audio processing method further includes:

[0018] Obtaining a reference timestamp according to a first timestamp of a target audio file that is too short and a second timestamp of a target audio file;

[0019] If the deviation between the merged timestamp of the merged audio file and the reference timestamp meets the preset condition, the merged audio file is taken as a valid audio file to perform self-optimization processing on the target audio file that is too short.

[0020] In one embodiment, when the quality assessment result is other than valid audio and invalid audio, the step of performing self-optimization processing on the corresponding audio file to be processed includes:

[0021] If the audio loudness value of the audio file to be processed does not fall within the preset coverage area of ​​the loudness normal distribution curve of the audio group, the corresponding quality assessment result is an assessment result other than valid audio and invalid audio, and the audio file to be processed is an abnormal loudness audio file;

[0022] Obtaining a loudness deviation between an abnormal audio loudness value of an audio file with abnormal loudness and a preset audio loudness value;

[0023] The audio loudness of the audio file with abnormal loudness is adjusted according to the smaller value between the loudness deviation and the adjustment limit value, so as to perform self-optimization processing on the audio file with abnormal loudness.

[0024] In one embodiment, the step of performing quality assessment on each audio file to be processed to obtain a quality assessment result of each audio file to be processed includes:

[0025] Dividing the plurality of audio files to be processed into a plurality of audio groups according to the number of channels and the sampling rate of each audio file to be processed;

[0026] For each audio group, the repairability score of any too short audio file is determined according to the continuity probability of any annotated text of any too short audio file in the audio group and the second concatenated text of the adjacent annotated text, wherein the too short audio file is an audio file whose audio duration is less than a preset duration and / or the annotated text length is less than a preset text length, and the adjacent annotated text is the annotated text of the adjacent audio file in the audio group whose subscript is adjacent to the subscript of any too short audio file, and the adjacent audio files include a first audio file whose subscript is greater than the subscript of any too short audio file and / or a second audio file whose subscript is less than the subscript of any too short audio file;

[0027] For each audio group, determine the too short audio ratio score of the audio group according to the ratio of too short audio files in the audio group;

[0028] For each audio group, a loudness score of the audio group is obtained according to a probability of an audio file in the audio group falling into a preset coverage area of ​​a loudness normal distribution curve of the audio group, wherein the loudness normal distribution curve is constructed according to an average loudness value and a loudness standard deviation of each audio file in the audio group;

[0029] For each audio group, according to whether the loudness difference between the abnormal audio loudness value of each audio file with abnormal loudness in the audio group and the preset audio loudness value is within a preset loudness difference range, determine the abnormal loudness score of each audio file with abnormal loudness;

[0030] For each audio group, the quality assessment result of the audio group is determined based on the repairability score, the proportion of too short audio, the loudness score and the abnormal loudness scores of the audio group.

[0031] In one embodiment, for each audio group, the step of determining the repairability score of any too short audio file according to the continuity probability of any annotated text of any too short audio file in the audio group and the concatenated text of the selected annotated text includes:

[0032] For each audio group, if the adjacent audio file of any too short audio file in the audio group is an too short audio file, the continuity probability of any annotated text of any too short audio file and the second concatenated text of the adjacent annotated text of the adjacent audio file is used as the repairability score of any too short audio file;

[0033] For each audio group, if the adjacent audio file of any too short audio file in the audio group is not an too short audio file, the repairability score is determined according to the continuity probability of the third spliced ​​text and / or the continuity probability of the fourth spliced ​​text, the third spliced ​​text is the spliced ​​text of any annotated text and the first annotated text of the first audio file, and the fourth spliced ​​text is the spliced ​​text of any annotated text and the second annotated text of the second audio file.

[0034] In addition, to achieve the above purpose, the present application also proposes an audio processing device, the audio processing device comprising:

[0035] An acquisition module, used to acquire multiple audio files to be processed;

[0036] An evaluation module is used to perform quality evaluation on each audio file to be processed and obtain a quality evaluation result of each audio file to be processed;

[0037] A processing module, used for treating the corresponding audio file to be processed as a valid audio file when the quality assessment result is valid audio;

[0038] The processing module is also used to delete the corresponding audio file to be processed when the quality assessment result is invalid audio;

[0039] The processing module is also used to perform self-optimization processing on the corresponding audio file to be processed when the quality assessment result is other assessment results other than valid audio and invalid audio.

[0040] In addition, to achieve the above objectives, the present application also proposes an audio processing device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the audio processing method as described above.

[0041] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the audio processing method as described above are implemented.

[0042] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the audio processing method as described above are implemented.

[0043] One or more technical solutions proposed in this application have at least the following technical effects:

[0044] The present application provides an audio processing method, apparatus, device, storage medium and program product. After obtaining quality assessment results of multiple audio files to be processed through evaluation, the audio files to be processed whose quality assessment results are valid audio are taken as valid audio files, the audio files to be processed whose quality assessment results are invalid audio are deleted, and the audio files to be processed whose quality assessment results are other assessment results except valid audio and invalid audio are liberalized, so that different processing measures are taken for the audio files to be processed with different quality assessment results. Compared with only deleting erroneous data and invalid data, the final data set retains valid data and other self-optimized data. Since the self-optimization processing is performed on the data other than the valid data and the invalid data, the quality of the entire data set is improved. When applied to the multimodal annotation data processing of a large model, the overall quality of the multimodal annotation data of the large model is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is a flowchart of the first embodiment of the audio processing method of the present application;

[0048] Figure 2 An exemplary annotated text example diagram provided for the audio processing method of the present application;

[0049] Figure 3 Another exemplary annotated text example diagram provided for the audio processing method of the present application;

[0050] Figure 4 An exemplary self-optimization result display interface diagram provided for the audio processing method of the present application;

[0051] Figure 5 This is a module diagram of the first embodiment of the audio processing device of the present application;

[0052] Figure 6 Schematic diagram of the device structure of the hardware operating environment involved in the audio processing method in the embodiment of the present application.

[0053] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0055] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0056] The main solution of the embodiment of the present application is: obtain multiple audio files to be processed; perform quality assessment on each audio file to be processed to obtain a quality assessment result of each audio file to be processed; when the quality assessment result is valid audio, treat the corresponding audio file to be processed as a valid audio file; when the quality assessment result is invalid audio, delete the corresponding audio file to be processed; when the quality assessment result is other assessment results except valid audio and invalid audio, perform self-optimization processing on the corresponding audio file to be processed.

[0057] Based on big data technology, data item business rules, data item keywords and data item association rules are used to evaluate the quality of multimodal annotated data of large models and obtain data quality scores. Among them, data with a data quality score higher than 90 points are usually high-value data, data with a data quality score lower than 60 points are usually erroneous data and invalid data, and data with a data quality score between 60 and 90 points are usually ordinary value data. In related technologies, in order to improve the quality of multimodal annotated data, erroneous data and invalid data are usually deleted. However, only deleting low-quality erroneous data and invalid data still cannot guarantee the overall quality of the data.

[0058] Therefore, the present application provides a solution. After evaluating and obtaining the quality assessment results of multiple audio files to be processed, the audio files to be processed whose quality assessment results are valid audio are taken as valid audio files, the audio files to be processed whose quality assessment results are invalid audio are deleted, and the audio files to be processed whose quality assessment results are other assessment results except valid audio and invalid audio are liberalized, so that different processing measures are taken for the audio files to be processed with different quality assessment results. Compared with only deleting erroneous data and invalid data, the final data set retains the valid data and other self-optimized data. Since the self-optimization processing is performed on other data except valid data and invalid data, the quality of the entire data set is improved. When applied to the multimodal annotation data processing of large models, the overall quality of the multimodal annotation data of the large model is effectively improved.

[0059] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an audio processing device, etc. The following takes an audio processing device as an example to illustrate this embodiment and the following embodiments.

[0060] Based on this, the present application embodiment provides an audio processing method, referring to Figure 1 , Figure 1 This is a flowchart diagram of the first embodiment of the audio processing method of the present application.

[0061] In this embodiment, the audio processing method may include steps S100 to S500:

[0062] Step S100, obtaining a plurality of audio files to be processed.

[0063] In this embodiment, the audio processing device can be a large model data supply platform. The data provider can upload the content of the audio file to be processed to the large model data supply platform through the web access interface; after the large model data supply platform performs necessary security checks and verifications on the received audio files to be processed, it can call the data quality monitoring service to perform quality assessment on the audio files to be processed; based on the quality assessment results, data processing is performed on the audio files to be processed, and the processing results are returned to the data provider in a visual manner via the web, and the data provider decides whether to accept the data processing results; finally, the processed audio files are stored in the warehouse for use by the large model, thereby ensuring the value of the data in the data supply platform.

[0064] It should be noted that if Figure 2 As shown, the audio file to be processed is annotated data, and the annotated text of the audio file to be processed may include the subscript Xid of the audio file to be processed and file description information, etc. For example, the annotated text of the audio file to be processed is "Weightlifting Championship Men's 67kg Final / 0063.wav", and the corresponding Xid is 0063.

[0065] Step S200: Perform quality assessment on each audio file to be processed to obtain a quality assessment result of each audio file to be processed.

[0066] It should be noted that for each audio file to be processed, a multi-faceted quality assessment can be performed to obtain corresponding quality assessment results in various aspects. For example, the quality of the audio file to be processed in terms of audio loudness and audio length can be evaluated. Specifically, for different usage scenarios, the audio loudness quality of the audio file to be processed can be evaluated based on whether the audio loudness of the audio file to be processed meets the corresponding preset loudness requirement. The audio length of the audio file to be processed can also be evaluated based on whether the audio length of the audio file to be processed meets the preset duration and / or whether the length of the annotated text meets the preset text length. Among them, the preset loudness requirement, preset duration and preset text length can all be set according to the actual usage scenario.

[0067] Step S300: When the quality assessment result is valid audio, the corresponding audio file to be processed is used as a valid audio file.

[0068] It should be noted that when the quality assessment result is valid audio, it means that the corresponding audio file to be processed can be a high-value audio file. The audio file to be processed is regarded as a valid audio file. The valid audio file does not need to be optimized and can be directly stored.

[0069] Step S400: When the quality assessment result is invalid audio, the corresponding audio file to be processed is deleted.

[0070] It should be noted that when the quality assessment result is invalid audio, it means that the corresponding audio file to be processed is an erroneous audio file and / or an invalid audio file, which has no use value for the large model. The audio file to be processed can be directly deleted without being put into storage.

[0071] Step S500: When the quality assessment result is other than valid audio and invalid audio, the corresponding audio file to be processed is self-optimized.

[0072] It should be noted that, when the quality assessment results are other than valid audio and invalid audio, it means that the corresponding audio file to be processed is an audio file of medium value or ordinary value. This type of audio file to be processed can be used by large models, but the use value is not high. At this time, the audio file to be processed can be self-optimized to improve the use value of the audio file to be processed.

[0073] It is understandable that the self-optimization process can be to optimize the quality of the audio file to be processed in terms of audio loudness and audio length, so that the audio loudness and audio length of the audio file to be processed meet the corresponding usage scenario requirements and the audio loudness requirements and audio length requirements of the large model data supply platform. For example, the audio loudness value of the audio file to be processed of medium value or ordinary value can be adjusted to a preset loudness range to optimize the quality of the audio file to be processed in terms of audio loudness; the audio file to be processed of medium value or ordinary value can be merged with the adjacent audio files with adjacent subscripts to optimize the quality of the audio file to be processed in terms of audio length.

[0074] Therefore, this embodiment provides an audio processing method. After obtaining the quality assessment results of multiple audio files to be processed through evaluation, the audio files to be processed whose quality assessment results are valid audio are taken as valid audio files, and the audio files to be processed whose quality assessment results are invalid audio are deleted, and the audio files to be processed whose quality assessment results are other assessment results except valid audio and invalid audio are liberalized, so that different processing measures are taken for the audio files to be processed with different quality assessment results. Compared with only deleting erroneous data and invalid data, the final data set retains valid data and other self-optimized data. Since the self-optimization processing is performed on other data except valid data and invalid data, the quality of the entire data set is improved. When applied to the processing of multimodal annotation data of large models, the overall quality of the multimodal annotation data of the large model is effectively improved.

[0075] In a feasible implementation, step S200 may include steps A210 to A260:

[0076] Step A210: divide the multiple audio files to be processed into multiple audio groups according to the number of channels and sampling rate of each audio file to be processed.

[0077] It should be noted that the sampling rate and the number of channels can jointly affect the quality and size of the audio file. Audio files with different sampling rates and numbers of channels can provide different sound quality and auditory experiences, and also require different storage space and processing capabilities. Therefore, according to the number of channels and sampling rates of each audio file to be processed, multiple audio files to be processed can be divided into multiple audio groups, and quality assessments can be performed on different audio groups to obtain the overall quality assessment results of the audio group, and the use value of the audio group can be determined based on the overall quality assessment results. For example, multiple audio files to be processed can be divided into audio groups such as mono-16K, mono-8K, dual-channel-16K, and dual-channel-44.1K.

[0078] It can be understood that the audio information of the audio file to be processed can be extracted from the audio file to be processed, wherein the audio information may include the number of channels, bit depth, sampling rate, audio loudness value, etc.

[0079] Step A220: For each audio group, a repairability score of any too short audio file is determined according to the continuity probability between any annotated text of any too short audio file in the audio group and the second concatenated text of the adjacent annotated text.

[0080] Among them, an audio file that is too short is an audio file whose audio duration is less than a preset duration and / or whose annotated text length is less than a preset text length, and the adjacent annotated text is the annotated text of an adjacent audio file in the audio group that is adjacent to the subscript number of any too short audio file, and the adjacent audio files include a first audio file whose subscript number is greater than the subscript number of any too short audio file and / or a second audio file whose subscript number is less than the subscript number of any too short audio file.

[0081] It should be noted that after a plurality of audio files to be processed are grouped, for each audio group, it is possible to determine whether each audio file is an overly short audio file based on the audio length and / or the length of the annotated text of each audio file in the audio group, and mark the audio files whose audio duration is less than the preset duration and / or the length of the annotated text is less than the preset text length as overly short audio files. Preferably, audio files with shorter audio durations in the audio group are preferentially marked as overly short audio files to avoid the situation where the audio duration is long but the effective annotated text length is very short. Among them, the preset duration can be set according to the actual usage scenario, for example, the preset duration can be 2s to 5s. Overly short audio files can be marked as TM_short, and the total number of overly short audio files can be expressed as size(TM_short).

[0082] It can be understood that the repairable score can represent the correlation between any too short audio file and the adjacent audio file. Since the content of the annotation text can correspond to the content of the audio file, the correlation between any too short audio file and the adjacent audio file can be determined according to the continuity probability of the second spliced ​​text, that is, the repairable score of any too short audio file can be determined according to the continuity probability of the second spliced ​​text. Among them, the number of adjacent audio files of any too short audio file in an audio group can be one or two; when the number of adjacent audio files of any too short audio file is one, the repairable score can be the continuity probability of any annotation text and the second spliced ​​text of the adjacent annotation text of the adjacent audio file; when the number of adjacent audio files of any too short audio file is two, the repairable score can be the continuity probability of any annotation text and the second spliced ​​text of the adjacent annotation text of the two adjacent audio files, or the repairable score can be the continuity probability of any annotation text and the second spliced ​​text of any adjacent annotation text of any adjacent audio file. The continuity probability of the second spliced ​​text can be calculated by NLP technology.

[0083] In actual use, when the continuity probability of the second spliced ​​text is greater than or equal to a preset threshold, the second spliced ​​text is considered to be a sentence, which can indicate that any audio file that is too short can be merged with an adjacent audio file, and the closer the repairability score of any audio file that is too short is to 1; when the continuity probability of the second spliced ​​text is less than the preset threshold, the second spliced ​​text is considered not to be a sentence, which can indicate that any audio file that is too short cannot be merged with an adjacent audio file, and the closer the repairability score of any audio file that is too short is to 0, the more likely it is that any audio file that is too short cannot be self-optimized.

[0084] Step A230: For each audio group, determine the too short audio ratio score of the audio group according to the ratio of the too short audio files in the audio group.

[0085] It should be noted that when evaluating the quality assessment results of the entire audio group, the impact of the proportion of too short audio files in the audio group on the quality assessment results can also be considered. Among them, the higher the proportion of too short audio files in the audio group, the lower the value of the audio group, and the lower the proportion of too short audio files in the audio group, the higher the value of the audio group.

[0086] It can be understood that when the proportion of audio files that are too short exceeds the preset proportion, the score of the proportion of audio files that are too short h(TM) can be a negative value; when the proportion of audio files that are too short is lower than the preset proportion, the score of the proportion of audio files that are too short h(TM) can be a positive value; wherein, the preset proportion can be set according to actual usage requirements, for example, the preset proportion can be 10%.

[0087] Step A240: For each audio group, a loudness score of the audio group is obtained according to the probability of audio files in the audio group falling into a preset coverage area of ​​a loudness normal distribution curve of the audio group.

[0088] The loudness normal distribution curve is constructed according to the loudness average and loudness standard deviation of each audio file in the audio group.

[0089] It should be noted that after the multiple audio files to be processed are grouped, for each audio group, it is also possible to determine whether each audio file is an audio file with abnormal loudness according to the audio loudness value of each audio file in the audio group, and mark the audio files with smaller audio loudness values ​​and / or larger audio loudness values ​​in each audio group as audio files with abnormal loudness.

[0090] In specific use, mathematical statistics are performed on the audio loudness values ​​of each audio file in the audio group, and after the loudness average, loudness variance and loudness standard deviation of the audio group are calculated, a corresponding normal distribution can be constructed based on the obtained loudness average and loudness standard deviation to obtain a corresponding loudness normal distribution curve; and the audio files that do not fall into the preset coverage area of ​​the loudness normal distribution curve are marked as loudness abnormal audio files. Among them, loudness abnormal audio files can include larger audio files and smaller audio files. Larger audio files can be marked as LDN_height, and smaller audio files can be marked as LDN_low. The number of audio files in larger audio files can be expressed as size(LDN_height), and the number of audio files in smaller audio files can be expressed as size(LDN_low). The preset coverage area can be set according to actual usage requirements. For example, the preset coverage area is the μ±σ area of ​​the loudness normal distribution curve.

[0091] It is understandable that, for an audio group, the probability of an audio file in the audio group falling into a preset coverage area of ​​a corresponding loudness normal distribution curve can be used as the loudness score f(LDN) of the audio group. The loudness score can reflect the number of normal loudness audio files other than abnormal loudness audio files in the corresponding audio group.

[0092] Step A250: for each audio group, according to whether the loudness difference between the abnormal audio loudness value of each audio file with abnormal loudness in the audio group and the preset audio loudness value is within the preset loudness difference range, determine the abnormal loudness score of each audio file with abnormal loudness.

[0093] It is understandable that the preset audio loudness value can be determined based on the audio loudness values ​​of existing audio files in the database of the large model data supply platform. The greater the loudness difference between the abnormal audio loudness value of the loudness-abnormal audio file and the preset audio loudness value, the lower the value of the loudness-abnormal audio file; the smaller the loudness difference between the abnormal audio loudness value of the loudness-abnormal audio file and the preset audio loudness value, the higher the value of the loudness-abnormal audio file.

[0094] In specific use, the abnormal loudness score g(LDN) can be a piecewise function. If the loudness difference between the abnormal audio loudness value of the loudness-abnormal audio file and the preset audio loudness value is within a preset loudness difference range, the abnormal loudness score of the loudness-abnormal audio file is a positive value; if the loudness difference between the abnormal audio loudness value of the loudness-abnormal audio file and the preset audio loudness value is higher than the maximum value of the preset loudness difference range, or lower than the minimum value of the preset loudness difference range, the abnormal loudness score of the loudness-abnormal audio file is a negative value.

[0095] Step A260: for each audio group, determine the quality assessment result of the audio group according to the repairability score, the too short audio ratio score, the loudness score and the abnormal loudness scores of the audio group.

[0096] It should be noted that, for an audio group, the quality assessment result may be the quality score of the audio group, and the quality score of the audio group may be calculated by combining the repairability scores, the too-short audio ratio scores, the loudness scores, and the abnormal loudness scores of the audio group.

[0097] The quality score Q of the audio group can be calculated using the following formula 1:

[0098] Q=f(LDN)*c1+g(LDN)*c2+h(TM)*c3+M

[0099] c1 is the loudness score weight of the audio group, c2 is the abnormal loudness score weight of the audio group, c3 is the short audio ratio score weight of the audio group, and M is the total repairability score of each short audio file in the audio group. The score weight can be set according to actual usage. For example, c1 can be 60, c2 can be 10, and c3 can be 50.

[0100] It is understandable that the large model data supply platform can feed back the quality assessment results of each audio group to the data provider, so that the data provider can perform quality optimization and other processing on the audio files in each audio group according to the quality assessment results.

[0101] Therefore, this embodiment provides an audio processing method, which performs quality evaluation of audio loudness and audio duration on audio files in each audio group according to the audio loudness distribution and the too-short audio distribution in the audio group, thereby performing multi-faceted quality evaluation of the audio in the audio group in combination with the actual distribution of the audio files in the audio group, which can improve the reliability of the quality evaluation results of the audio group and improve the evaluation accuracy of each audio group.

[0102] In a feasible implementation, step A220 may include: for each audio group, if the adjacent audio file of any too short audio file in the audio group is an too short audio file, then the continuity probability of any annotated text of any too short audio file and the second concatenated text of the adjacent annotated text of the adjacent audio file is used as the repairability score of any too short audio file; for each audio group, if the adjacent audio file of any too short audio file in the audio group is not an too short audio file, then the repairability score is determined according to the continuity probability of the third concatenated text and / or the continuity probability of the fourth concatenated text.

[0103] The third concatenated text is a concatenated text of any annotation text and a first annotation text of the first audio file, and the fourth concatenated text is a concatenated text of any annotation text and a second annotation text of the second audio file.

[0104] It should be noted that, for any audio group, the adjacent audio files of any too short audio file in the audio group are too short audio files, including: the first audio file is too short audio file, the second audio file is too short audio file, and the first audio file and the second audio file are both too short audio files; correspondingly, the second spliced ​​text is the spliced ​​text of any annotation text and the first annotation text, or the spliced ​​text of any annotation text and the second annotation text, or the spliced ​​text of any annotation text and the first annotation text and the second annotation text.

[0105] For any audio group, the adjacent audio files of any too short audio file in the audio group are not too short audio files, including: the number of adjacent audio files of any too short audio file is one, and the adjacent audio file is not too short audio file, and the number of adjacent audio files of any too short audio file is two, and both adjacent audio files are not too short audio files; correspondingly, the repairability score can be the larger of the continuity probability of the third spliced ​​text, the continuity probability of the fourth spliced ​​text, or the continuity probability of the third spliced ​​text and the continuity probability of the fourth spliced ​​text.

[0106] Therefore, this embodiment provides an audio processing method, which determines the repairability score of an audio file that is too short according to the continuity probability of different spliced ​​texts of the annotated text of the audio file that is too short and the adjacent annotated text of an adjacent audio file, thereby determining the repairability score of the audio file that is too short based on multiple repairable situations of the audio file that is too short, thereby improving the accuracy of the repairability score.

[0107] In a feasible implementation, the multiple audio files to be processed include short audio files whose audio duration is less than a preset duration and / or whose annotated text length is less than a preset text length, and the quality assessment result includes a repairability score of the short audio files. Step S500 may include steps A510 to A550:

[0108] Step A510, concatenating the target annotated text of the target overshort audio file whose repairability score is greater than or equal to a preset repairability score with the selected annotated text of the selected audio file to obtain a first concatenated text.

[0109] Among them, when the repairability score is greater than or equal to the preset repairability score, the quality assessment results are other assessment results except for valid audio and invalid audio, and the selected audio files include at least one to-be-processed audio file adjacent to the subscript number of the target too short audio file in the audio group where the target too short audio file is located.

[0110] It should be noted that the short audio files with a repairability score less than the preset repairability score are erroneous audio files and / or invalid audio files, while the target short audio files with a repairability score greater than or equal to the preset repairability score are audio files of medium value or ordinary value. For such target short audio files, their use value can be improved through self-optimization processing. The preset repairability score can be set according to actual usage, for example, the preset repairability score can be 0.6.

[0111] It is understandable that the selected audio files may include at least one to-be-processed audio file before the subscript number of the target too short audio file in the audio group where the target too short audio file is located and / or at least one to-be-processed audio file after the subscript number of the target too short audio file in the audio group. Preferably, the selected audio files include 3 to-be-processed audio files before the subscript number of the target too short audio file in the audio group where the target too short audio file is located and / or 3 to-be-processed audio files after the subscript number of the target too short audio file in the audio group.

[0112] Step A520: determining a target concatenated text where the target annotated text is located from the first concatenated text according to the punctuation positions in the first concatenated text.

[0113] It should be noted that the first concatenated text may include multiple sentences, and the multiple sentences may be separated by punctuation marks. The sentence including the target annotated text among the multiple sentences is the target concatenated text.

[0114] Step A530, determining a first selected concatenated text and a second selected concatenated text according to the position of the target annotated text in the target concatenated text and the target concatenated text.

[0115] It should be noted that the target concatenated text may include the target marked text and the marked text of at least one to-be-processed audio file before the subscript number of the target too-short audio file and / or the marked text of at least one to-be-processed audio file after the subscript number of the target too-short audio file in the audio group. Correspondingly, the first selected concatenated text may include the target marked text and the marked text of at least one to-be-processed audio file before the subscript number of the target too-short audio file, and the second selected concatenated text may include the target marked text and the marked text of at least one to-be-processed audio file after the subscript number of the target too-short audio file.

[0116] Step A540: Determine a target audio file from the selected audio files according to the first completeness probability of the first selected concatenated text and the second completeness probability of the second selected concatenated text.

[0117] It should be noted that, when the target splicing text includes the target annotated text and the annotated text of at least one audio file to be processed before the target too-short audio file subscript number, the target audio file includes at least one audio file to be processed before the target too-short audio file subscript number corresponding to the first selected splicing text.

[0118] When the target concatenated text includes the target annotated text and the annotated text of at least one audio file to be processed after the target too-short audio file subscript number, the target audio file includes at least one audio file to be processed after the target too-short audio file subscript number corresponding to the second selected concatenated text.

[0119] When the target concatenated text includes the target annotated text and the annotated text of at least one to-be-processed audio file before the subscript number of the target too-short audio file and the annotated text of at least one to-be-processed audio file after the subscript number of the target too-short audio file in the audio group, the target audio file includes at least one to-be-processed audio file corresponding to the selected concatenated text with a higher completeness probability. The completeness probability can be calculated by NLP technology.

[0120] Step A550, merging the target audio file that is too short with the target audio file to obtain a merged audio file, so as to perform self-optimization processing on the target audio file that is too short.

[0121] It should be noted that this embodiment combines the target too short audio file with the target audio file, thereby merging the too short audio file that can be repaired more easily into other audio files, reducing the existence of too short audio files, and completing the self-optimization processing of the target too short audio file.

[0122] In one example, if Figure 3 As shown, the first spliced ​​text is "So he must + use the power of his right leg to support + the body a little bit, and then raise his left leg again + and the two + legs are alternately used to exert force, + which still has a great influence on his performance.", and the corresponding target spliced ​​text is "Then the left leg is raised again and the two legs are alternately used to exert force", so the first selected spliced ​​text is Text_fup = "Then the left leg is raised again", and the second selected spliced ​​text is Text_fdown = "The two legs are alternately used to exert force". Through NLP, it can be calculated that the integrity probability of Text_fdown is higher than the integrity probability of Text_fup, so that the audio file of sequence number 67 can be merged downward with the audio file of sequence number 68 to complete the self-optimization processing of the audio file of sequence number 67.

[0123] Therefore, this embodiment provides an audio processing method, which realizes self-optimization processing of short audio files by merging short audio files with strong repairability into target audio files. And by comparing the integrity probability of the spliced ​​text corresponding to different merging methods, the merging strategy of short audio files with strong repairability is determined, thereby improving the accuracy of self-optimization processing of short audio files.

[0124] In a feasible implementation, after step A550, the audio processing method may further include: obtaining a reference timestamp based on the first timestamp of the target audio file that is too short and the second timestamp of the target audio file; if the deviation between the merged timestamp of the merged audio file and the reference timestamp meets a preset condition, the merged audio file is used as a valid audio file to perform self-optimization processing on the target audio file that is too short.

[0125] It should be noted that the first timestamp and the second timestamp can be determined according to the end and start fields in the annotation text of the corresponding audio file. The merged timestamp can be obtained by inputting the merged audio file into the VAD model for inference.

[0126] If the deviation between the merge timestamp and the reference timestamp is not large, it means that the repair of the target audio file is effective and the merged audio file is a valid audio file; if the deviation between the merge timestamp and the reference timestamp is large, it means that the repair of the target audio file is invalid and the merge is rolled back without merging the target audio file with the target audio file. The preset condition may be a preset deviation range.

[0127] Therefore, this embodiment provides an audio processing method, which determines whether the repair of the target too-short audio file is effective by judging whether the deviation between the merged timestamp of the merged audio file and the reference timestamp meets the preset conditions, thereby realizing the self-optimization result verification of the too-short audio file and improving the self-optimization accuracy of the too-short audio file.

[0128] In a feasible implementation, step S500 may include: if the audio loudness value of the audio file to be processed does not fall within the preset coverage area of ​​the loudness normal distribution curve of the audio group, the corresponding quality assessment result is an assessment result other than valid audio and invalid audio, and the audio file to be processed is an abnormal loudness audio file; obtaining a loudness deviation between the abnormal audio loudness value of the audio file with abnormal loudness and a preset audio loudness value; and adjusting the audio loudness of the audio file with abnormal loudness according to the smaller value between the loudness deviation and the adjustment limit value, so as to perform self-optimization processing on the audio file with abnormal loudness.

[0129] It should be noted that if the audio loudness value does not fall into the preset coverage area of ​​the loudness normal distribution curve of the audio group, it means that the audio file to be processed is an audio file with abnormal loudness. For audio files with abnormal loudness, the audio loudness of the audio file with abnormal loudness can be fine-tuned according to the smaller value between the loudness deviation and the adjustment limit value, so as to achieve self-optimization processing of the audio file with abnormal loudness. Among them, the adjustment limit value can be set according to the actual use requirements of the large model data supply platform.

[0130] It is understandable that for each audio group, after the quality of each audio group is evaluated, the target short audio files and loudness abnormal audio files in the audio group can also be self-optimized according to the self-optimization method provided in this embodiment to obtain the self-optimized audio group. The large model data supply platform can output each audio group and each optimized audio group to the human-computer interaction system, and the self-optimization results can be displayed in groups through the human-computer interaction system. And the audio files are organized in a foldable list through the human-computer interaction system.

[0131] like Figure 4 As shown, users can select a specific optimization data in a certain optimization category group to compare before and after optimization. The comparison results will be displayed on the left, in two parts. Among them, the optimization category group corresponds to different audio groups, the original data is displayed as the audio file before optimization, and the optimized data is displayed as the audio file after optimization.

[0132] Through the human-computer interaction system, when the user clicks the "Discard" button, the optimization category group is discarded as a whole and the optimization results are not adopted; when the user clicks the "Pass" button, the optimization category group is passed as a whole and the optimization results are adopted. By comparing the optimization results before and after, users can choose whether to accept the optimization results in a more fine-grained manner. "Single submission" is suitable for abandoning most of the optimization category groups and only adopting a few optimization results; "Remove from group" is suitable for adopting most of the optimization category groups and only abandoning a few optimization results.

[0133] Through the interaction of the human-computer interaction system, the underlying data processing can be triggered, so that the adopted optimization data is uploaded to the "retention area" of the large model data supply platform, waiting for the back-end staff or data users to determine whether to use the adopted optimization data, while the unadopted optimization data is traced back to the data provider.

[0134] Therefore, this embodiment provides an audio processing method, which implements self-optimization processing of audio files with abnormal loudness by fine-tuning the audio loudness of audio files with abnormal loudness according to the smaller value between the loudness deviation and the adjustment limit value, thereby preventing the audio files from having poor loudness and affecting their use value. In addition, by fine-tuning the audio loudness of audio files with abnormal loudness in combination with the adjustment limit value, popping protection processing is implemented, ensuring that popping will not occur in the self-optimized audio files when they are played.

[0135] The present application provides an audio processing device, such as Figure 5 As shown, the audio processing device may include:

[0136] An acquisition module 10 is used to acquire a plurality of audio files to be processed;

[0137] An evaluation module 20 is used to perform quality evaluation on each audio file to be processed to obtain a quality evaluation result of each audio file to be processed;

[0138] A processing module 30, configured to, when the quality assessment result is valid audio, use the corresponding audio file to be processed as a valid audio file;

[0139] The processing module 30 is further used to delete the corresponding audio file to be processed when the quality assessment result is invalid audio;

[0140] The processing module 30 is further used to perform self-optimization processing on the corresponding audio file to be processed when the quality assessment result is other assessment results except valid audio and invalid audio.

[0141] For more implementation details in the specific implementation of the above audio processing device, please refer to the description of the specific implementation of the audio processing method in the above embodiment. For the sake of brevity of the description, they will not be repeated here.

[0142] The present application provides an audio processing device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the audio processing method in the above-mentioned embodiment 1.

[0143] Reference below Figure 6 , which shows a schematic diagram of the structure of an audio processing device suitable for implementing an embodiment of the present application. The audio processing device in the embodiment of the present application may include but is not limited to mobile terminals such as laptop computers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), etc., and fixed terminals such as desktop computers, etc. Figure 6 The audio processing device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0144] like Figure 6 As shown, the audio processing device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the audio processing device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the audio processing device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an audio processing device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0145] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0146] The audio processing device provided by the present application adopts the audio processing method in the above embodiment, which can solve the technical problem of how to improve the overall quality of multimodal annotation data of large models. Compared with the related art, the beneficial effects of the audio processing device provided by the present application are the same as the beneficial effects of the audio processing method provided by the above embodiment, and the other technical features in the audio processing device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0147] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0148] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0149] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the audio processing method in the above-mentioned embodiment.

[0150] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0151] The computer-readable storage medium may be included in the audio processing device; or may exist independently without being assembled into the audio processing device.

[0152] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the audio processing device, the audio processing device: obtains multiple audio files to be processed; performs quality assessment on each audio file to be processed to obtain a quality assessment result of each audio file to be processed; when the quality assessment result is valid audio, the corresponding audio file to be processed is a valid audio file; when the quality assessment result is invalid audio, the corresponding audio file to be processed is deleted; when the quality assessment result is other assessment results except valid audio and invalid audio, the corresponding audio file to be processed is self-optimized.

[0153] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0154] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0155] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0156] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned audio processing method, and can solve the technical problem of how to improve the overall quality of multimodal annotation data of large models. Compared with the related art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the audio processing method provided in the above-mentioned embodiment, and will not be repeated here.

[0157] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned audio processing method when executed by a processor.

[0158] The computer program product provided by this application can solve the technical problem of how to improve the overall quality of multimodal annotation data of large models. Compared with the relevant technology, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the audio processing method provided by the above embodiment, which will not be repeated here.

[0159] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. An audio processing method, characterized in that: The audio processing method comprises: Get multiple audio files to be processed; Performing quality assessment on each of the audio files to be processed to obtain a quality assessment result of each of the audio files to be processed; When the quality assessment result is valid audio, the corresponding audio file to be processed is used as a valid audio file; If the quality assessment result is invalid audio, the corresponding audio file to be processed is deleted; When the quality assessment result is other assessment results except the valid audio and the invalid audio, the corresponding audio file to be processed is self-optimized.

2. The audio processing method according to claim 1, characterized in that: The multiple audio files to be processed include short audio files whose audio duration is less than a preset duration and / or whose annotated text length is less than a preset text length, and the quality assessment result includes a repairability score of the short audio files; When the quality assessment result is other assessment results other than the valid audio and the invalid audio, the step of performing self-optimization processing on the corresponding audio file to be processed comprises: splicing the target annotated text of the target too short audio file whose repairability score is greater than or equal to the preset repairability score with the selected annotated text of the selected audio file to obtain a first spliced ​​text, wherein, when the repairability score is greater than or equal to the preset repairability score, the quality assessment result is other assessment results except the valid audio and the invalid audio, and the selected audio file includes at least one to-be-processed audio file in the audio group where the target too short audio file is located and whose subscript number is adjacent to the target too short audio file; Determining a target spliced ​​text where the target annotated text is located from the first spliced ​​text according to the punctuation positions in the first spliced ​​text; Determining a first selected spliced ​​text and a second selected spliced ​​text according to the position of the target annotated text in the target spliced ​​text and the target spliced ​​text; Determining a target audio file from the selected audio files according to a first integrity probability of the first selected concatenated text and a second integrity probability of the second selected concatenated text; The target too-short audio file is merged with the target audio file to obtain a merged audio file, so as to perform self-optimization processing on the target too-short audio file.

3. The audio processing method according to claim 2, characterized in that: After the step of merging the target too-short audio file with the target audio file to obtain a merged audio file, the audio processing method further includes: Obtaining a reference timestamp according to the first timestamp of the target too-short audio file and the second timestamp of the target audio file; If the deviation between the merged timestamp of the merged audio file and the reference timestamp meets a preset condition, the merged audio file is used as the valid audio file to perform self-optimization processing on the target too-short audio file.

4. The audio processing method according to claim 1, characterized in that: When the quality assessment result is other assessment results other than the valid audio and the invalid audio, the step of performing self-optimization processing on the corresponding audio file to be processed comprises: If the audio loudness value of the audio file to be processed does not fall within the preset coverage area of ​​the loudness normal distribution curve of the audio group, the corresponding quality assessment result is another assessment result other than the valid audio and the invalid audio, and the audio file to be processed is an abnormal loudness audio file; Obtaining a loudness deviation between an abnormal audio loudness value of the abnormal audio file and a preset audio loudness value; The audio loudness of the audio file with abnormal loudness is adjusted according to the smaller value between the loudness deviation and the adjustment limit value, so as to perform self-optimization processing on the audio file with abnormal loudness.

5. The audio processing method according to claim 1, characterized in that: The step of performing quality assessment on each of the audio files to be processed to obtain a quality assessment result of each of the audio files to be processed comprises: Dividing the plurality of audio files to be processed into a plurality of audio groups according to the number of channels and the sampling rate of each of the audio files to be processed; For each of the audio groups, the repairability score of any too short audio file in the audio group is determined according to the continuity probability of any annotated text of any too short audio file and the second concatenated text of the adjacent annotated text, wherein the too short audio file is an audio file whose audio duration is less than a preset duration and / or whose annotated text length is less than a preset text length, and the adjacent annotated text is the annotated text of an adjacent audio file in the audio group whose subscript is adjacent to the subscript of any too short audio file, and the adjacent audio files include a first audio file whose subscript is greater than the subscript of any too short audio file and / or a second audio file whose subscript is less than the subscript of any too short audio file; For each of the audio groups, determining a score for the proportion of too short audio files in the audio group according to the proportion of the too short audio files in the audio group; For each of the audio groups, a loudness score of the audio group is obtained according to a probability of an audio file in the audio group falling into a preset coverage area of ​​a loudness normal distribution curve of the audio group, wherein the loudness normal distribution curve is constructed according to a loudness average value and a loudness standard deviation of each of the audio files in the audio group; For each of the audio groups, determining an abnormal loudness score of each of the audio files with abnormal loudness in the audio group according to whether a loudness difference between an abnormal audio loudness value of the audio file with abnormal loudness and a preset audio loudness value is within a preset loudness difference range; For each of the audio groups, a quality assessment result of the audio group is determined according to the repairability scores, the too-short audio ratio scores, the loudness scores, and the abnormal loudness scores of the audio group.

6. The audio processing method according to claim 5, characterized in that: The step of determining the repairability score of any too short audio file in the audio group according to the continuity probability of any annotated text of any too short audio file in the audio group and the concatenated text of the selected annotated text comprises: For each of the audio groups, if an adjacent audio file of any too short audio file in the audio group is the too short audio file, a continuity probability between any annotated text of the too short audio file and a second concatenated text of an adjacent annotated text of the adjacent audio file is used as a repairability score of the too short audio file; For each of the audio groups, if the adjacent audio file of any of the too short audio files in the audio group is not the too short audio file, the repairability score is determined according to the continuity probability of a third spliced ​​text and / or the continuity probability of a fourth spliced ​​text, the third spliced ​​text being a spliced ​​text of any of the annotated texts and a first annotated text of the first audio file, and the fourth spliced ​​text being a spliced ​​text of any of the annotated texts and a second annotated text of the second audio file.

7. An audio processing device, characterized in that: The audio processing device comprises: An acquisition module, used to acquire multiple audio files to be processed; An evaluation module, used to perform quality evaluation on each of the audio files to be processed to obtain a quality evaluation result of each of the audio files to be processed; A processing module, configured to, if the quality assessment result is valid audio, use the corresponding audio file to be processed as a valid audio file; The processing module is further configured to delete the corresponding audio file to be processed when the quality assessment result is invalid audio; The processing module is further configured to perform self-optimization processing on the corresponding audio file to be processed when the quality assessment result is other assessment results other than the valid audio and the invalid audio.

8. An audio processing device, characterized in that: The audio processing device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the audio processing method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the audio processing method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the audio processing method according to any one of claims 1 to 6 are implemented.

Images