A data processing method and system for call voice translation

Abstract

The embodiments of the present application relate to the field of communication technology, and disclose a data processing method and system for call speech translation. The method selects different data processing modules according to the complexity of pre-processed texts to avoid a single data processing unit performing data processing on all translation requests at the same time, resulting in insufficient computing power and reduced translation efficiency and translation accuracy. For pre-processed texts with low complexity, a lightweight data processing module is selected for data processing, and there is no need to perform error detection, correction and other procedures, which can improve the efficiency of call speech translation. For pre-processed texts with high complexity, a high-level data processing module is selected for data processing, and error detection, correction and other procedures are performed to improve the accuracy of call speech translation. Therefore, the data processing method for call speech translation of the present application can meet the translation needs of different voice calls.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a data processing method and system for call speech translation. Background Art

[0002] As the pace of trade globalization accelerates, there is a need for barrier-free communication in certain call scenarios. For example, in a cross-language call scenario, a language translation function may be needed to help users understand the language of the other terminal user. In related technologies, when a communication server or a communication transfer station receives multiple translation requests at the same time, there is a phenomenon of insufficient computing power, resulting in reduced translation efficiency and translation accuracy, which has a certain impact on terminal users.

[0003] In the relevant patent technologies, such as the Chinese patent applications with patent numbers CN202211082451.5 and CN202311176267.1, although they disclose voice translation between different terminals to achieve barrier-free calls, they do not involve solving the problems of translation efficiency and accuracy. Summary of the invention

[0004] The main purpose of the present invention is to provide a data processing method and system for call voice translation, aiming to solve the technical problem in the prior art that when a communication server or a communication transfer station receives multiple translation requests at the same time, there is insufficient computing power resulting in reduced translation efficiency and translation accuracy.

[0005] To achieve the above-mentioned purpose, in a first aspect, an embodiment of the present application provides a data processing method for call speech translation, the method comprising:

[0006] Acquire voice and audio data sent by the first terminal, wherein the voice and audio data is voice and audio data in a first target language;

[0007] According to preset translation configuration information, the voice audio data sent by the first terminal is translated to obtain preprocessed text information in a second target language;

[0008] Determining the text length and text relevance of the preprocessed text according to the preprocessed text information;

[0009] Determining a redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text, wherein the redundancy coefficient is used to characterize the complexity of the preprocessed text;

[0010] According to the redundancy coefficient of the pre-processed text, a target data processing module is selected to perform text processing on the pre-processed text information to obtain target text information, wherein the target data processing module includes a lightweight data processing module and a high-level data processing module, and the complexity of the text processed by the high-level data processing module is greater than the complexity of the text processed by the lightweight data processing module;

[0011] Converting the target text information into speech audio data in a second target language;

[0012] The voice audio data in the second target language is sent to the second terminal according to the preset voice transmission configuration information.

[0013] In a possible implementation, the translation configuration information includes language configuration information, and the translating the voice audio data sent by the first terminal to obtain preprocessed text information in a second target language according to the preset translation configuration information includes:

[0014] According to preset language configuration information, the voice audio data sent by the first terminal is translated to obtain preprocessed text information in a second target language, wherein the language configuration information is used to represent the translation of the first target language into the second target language, or the translation of the second target language into the first target language.

[0015] In a possible implementation manner, before determining the text length and text relevance of the preprocessed text according to the preprocessed text information, the method further includes:

[0016] detecting whether there is text noise in the preprocessed text information in the second target language;

[0017] If text noise exists in the preprocessed text information in the second target language, the text noise in the preprocessed text information in the second target language is filtered.

[0018] In a possible implementation, determining the text length and text relevance of the preprocessed text according to the preprocessed text information includes:

[0019] Calling a text length function to perform text length identification on the preprocessed text information to obtain the text length of the preprocessed text;

[0020] The preprocessed text is input into a deep learning model to estimate the text relevance to obtain the text relevance of the preprocessed text.

[0021] In a possible implementation, determining the redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text includes:

[0022] The text length and text relevance of the preprocessed text are input into a redundant prediction model to obtain a redundant coefficient value, wherein the redundant prediction model satisfies the following expression:

[0023] C=(L / L0+1)*Ф1+(G / G0+1)*Ф2 (G≥0.5,L≤20 bytes)

[0024] C=(L / L0+1)*Ф2+(G / G0+1)*Ф1 (G<0.5,L>20 bytes)

[0025] In the formula, L0 is the length of the benchmark text, G0 is the relevance of the benchmark text, L is the text length of the preprocessed text, G is the text relevance of the preprocessed text, Ф1 and Ф2 are the corresponding influence weights, and Ф2>Ф1.

[0026] In a possible implementation, the step of selecting a target data processing module according to the redundancy coefficient of the preprocessed text to perform text processing on the preprocessed text information to obtain target text information includes:

[0027] Determine that the redundancy coefficient of the preprocessed text is greater than or equal to the redundancy coefficient threshold, select a high-level data processing module to perform text processing on the preprocessed text information to obtain target text information, and the high-level data processing module corrects language element errors in the text during the text processing to obtain the target text information;

[0028] Determine that the redundancy coefficient of the preprocessed text is less than a redundancy coefficient threshold, select a lightweight data processing module to perform text processing on the preprocessed text information to obtain target text information, and the lightweight data processing module identifies and segments text pauses during the text processing to obtain target text information.

[0029] In a possible implementation, the high-level data processing module includes a first language element error correction model, and the selecting high-level data processing module performs text processing on the preprocessed text information to obtain target text information, including:

[0030] detecting whether the preprocessed text has a first language element error;

[0031] It is determined that a first language element error exists in the preprocessed text, and the first language element error correction model is used to correct the first language element error in the preprocessed text to obtain target text information, wherein the first language element error correction model is an error correction model obtained by artificial intelligence machine training based on user historical feedback data.

[0032] In a possible implementation, the high-level data processing module includes a second language element error correction model, and the selecting high-level data processing module performs text processing on the preprocessed text information to obtain target text information, including:

[0033] detecting whether the preprocessed text has a second language element error;

[0034] It is determined that there are second language element errors in the preprocessed text, and the second language element error correction model is used to correct the second language element errors in the preprocessed text to obtain target text information, wherein the second language element error correction model is an error correction model obtained by artificial intelligence machine training based on user historical feedback data.

[0035] In a possible implementation, the lightweight data processing module includes a separator insertion algorithm, and the selecting lightweight data processing module to perform text processing on the preprocessed text information to obtain target text information includes:

[0036] According to a preset recognition rule, feature recognition is performed on the preprocessed text to obtain a segmentation boundary;

[0037] A separator insertion operation is performed before or after the segmentation boundary to obtain target text information.

[0038] In a possible implementation, converting the target text information into speech audio data in a second target language includes:

[0039] Determine the speech pause time interval according to a mapping table of the relationship between the spacer type and the time;

[0040] According to the voice pause time interval, a voice pause of a corresponding time is performed during the text conversion process on both sides of the separator.

[0041] In a possible implementation, the voice transmission configuration information includes one or more of speech speed configuration information, sound volume configuration information, and voice character object configuration information, and sending the voice audio data in the second target language to the second terminal according to the preset voice transmission configuration information includes:

[0042] Determine the target speech speed, target sound volume and target voice character object according to the speech speed configuration information, the sound volume configuration information and the voice character object configuration information;

[0043] The speech audio data of the second target language is sent to the second terminal according to one or more of the target speech speed, the target sound volume and the target speech character object.

[0044] In a second aspect, an embodiment of the present application also provides a data processing system for call speech translation, comprising: a memory, the memory being used to store program code; and a processor, the processor being used to call the program code to execute the method described in the first aspect.

[0045] Different from the prior art, the data processing method for call speech translation provided in the embodiment of the present application first obtains the speech and audio data sent by the first terminal; then, according to the preset translation configuration information, translates the speech and audio data sent by the first terminal to obtain preprocessed text information in the second target language; then, according to the preprocessed text information, determines the text length and text relevance of the preprocessed text; then, according to the text length and text relevance of the preprocessed text, determines the redundancy coefficient of the preprocessed text, and the redundancy coefficient is used to characterize the complexity of the preprocessed text; then, according to the redundancy coefficient of the preprocessed text, selects the target data processing module to perform text processing on the preprocessed text information to obtain the target text information; then, converts the target text information into speech and audio data in the second target language; finally, according to the preset speech transmission configuration information, sends the speech and audio data in the second target language to the second terminal.

[0046] Further, when the redundancy coefficient of the preprocessed text is greater than or equal to the redundancy coefficient threshold, a high-level data processing module is selected to perform text processing on the preprocessed text information to obtain the target text information; when the redundancy coefficient of the preprocessed text is less than the redundancy coefficient threshold, a lightweight data processing module is selected to perform text processing on the preprocessed text information to obtain the target text information. In this way, different data processing modules are selected according to the complexity of the preprocessed text to avoid a single data processing unit from performing data processing on all translation requests at the same time, resulting in insufficient computing power and reduced translation efficiency and translation accuracy. For preprocessed texts with low complexity, a lightweight data processing module is selected for data processing, and no error detection, correction and other procedures are required, which can improve the efficiency of call voice translation; and for preprocessed texts with high complexity, a high-level data processing module is selected for data processing, and error detection, correction and other procedures are required, which can improve the accuracy of call voice translation. Therefore, the data processing method for call voice translation of the present application can simultaneously translate and process voice data sent by different terminals to meet the translation needs of different voice calls. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art 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 the structures shown in these drawings without paying creative work.

[0048] Figure 1 A flowchart of a data processing method for call speech translation in some embodiments of the present application;

[0049] Figure 2 A flowchart of a data processing method for call speech translation in other embodiments of the present application;

[0050] Figure 3 This is a schematic diagram of the hardware structure of a data processing system for call speech translation in some embodiments of the present application.

[0051] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0053] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0054] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0055] As the pace of trade globalization accelerates, there is a need for barrier-free communication in certain call scenarios. For example, in a cross-language call scenario, a language translation function may be needed to help users understand the language of the other terminal user. In related technologies, when a communication server or a communication transfer station receives multiple translation requests at the same time, there is a phenomenon of insufficient computing power, resulting in reduced translation efficiency and translation accuracy, which has a certain impact on terminal users.

[0056] In view of the above problems, this application proposes a data processing method for call voice translation, which is applied to a communication server or a communication transfer station. When the communication server or the communication transfer station is used as the execution subject of the method of this application, the voice data is successively received, translated into text, processed into text, converted into voice, and sent, so as to realize barrier-free calls on different terminals. Figure 1-2 As shown, the following will mainly describe the specific steps of the data processing method for call voice translation. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order from here. Figure 1 , the method comprises the following steps:

[0057] S100, acquiring voice and audio data sent by a first terminal, wherein the voice and audio data is voice and audio data in a first target language;

[0058] The first terminal can be a mobile terminal device such as a mobile phone or a tablet, and the second terminal can also be a mobile terminal device such as a mobile phone or a tablet. The function of the communication server or the communication transfer station is to send the signal transmitted by the mobile phone to the communication network, and then transmit it to the target mobile phone or other device. In the voice call translation service, the communication server or the communication transfer station also translates and processes the voice data to achieve barrier-free communication between different terminals.

[0059] The first target language corresponds to the second target language described below, and can be any language that can be translated and configured by the communication server or the communication transfer station. This article takes English as the first target language and Chinese as the second target language as an example for explanation.

[0060] S200: translating the voice audio data sent by the first terminal according to preset translation configuration information to obtain preprocessed text information in a second target language;

[0061] The translation configuration information may be a translation speed configuration or a translation language configuration. The translation language configuration may be a translation configuration between a first target language and a second target language, such as a translation configuration between Chinese and English. After the translation configuration information is determined, the system can only translate between the first target language (English) and the second target language (Chinese), that is, it can only translate the first target language (English) into the second target language (Chinese), or translate the second target language (Chinese) into the first target language (English).

[0062] After receiving the voice and audio data sent by the first terminal, the communication server or the communication transfer station translates the voice and audio data sent by the first terminal to obtain pre-processed text information in a second target language.

[0063] For example, after receiving the voice data "we don't need to go to school today If the storm is coming", the text information "we don't need to go to school today if the storm is coming" or "if the storm is coming, we don't need to go to school today" is obtained through translation.

[0064] Since the text information obtained by preliminary translation may have certain defects and needs to be processed, the text information obtained by translation is referred to as pre-processed text information in this application.

[0065] It is understandable that due to the different grammatical rules between different languages, there may be differences in the order of translated sentences, and the more complex the sentence, the greater the probability that the sentence order does not conform to the rules of the target language. For example, the above-mentioned "we don't need to go to school today If the storm is coming" may have two translation results: "We don't need to go to school today if the storm is coming" or "If the storm is coming, we don't need to go to school today"; the simpler the sentence, the lower the probability of translation errors. Therefore, in order to improve the timeliness and accuracy of translation, different translation resources may be required for speech sentences of different complexity. For the speech of complex sentences, complex algorithms can be used for translation processing, and for the speech of simple sentences, simple algorithms can be used for translation processing; in this way, the efficiency of translation can be improved.

[0066] S300, determining the text length and text relevance of the preprocessed text according to the preprocessed text information;

[0067] In order to prevent noise from affecting the accuracy of text parameter collection of the preprocessed text, such as affecting the accuracy of text length and text relevance collection, in one embodiment, before the step S300: determining the text length and text relevance of the preprocessed text according to the preprocessed text information, it also includes:

[0068] detecting whether there is text noise in the preprocessed text information in the second target language;

[0069] If text noise exists in the preprocessed text information in the second target language, the text noise in the preprocessed text information in the second target language is filtered.

[0070] Specifically, text noise may be text noise generated by the voice noise generated by the external environment when the first terminal is making a call, which is obtained by the communication server or the communication transfer station and then translated into text noise. For example, when someone else speaks, the voice that is not related to the current call is recognized and then translated into text information. This part of the noise is text noise; text noise can be judged and identified based on the volume of the main voice and the noise. Generally speaking, the noise is smaller. Therefore, when it is judged that there is text noise in the preprocessed text information of the second target language, the text noise in the preprocessed text information of the second target language is filtered to improve the accuracy of text parameter collection.

[0071] The complexity of the text after speech translation is related to many factors, such as the text length and the correlation between the text context mentioned above. The longer the text is, the more complex the text is. The greater the correlation of the text is, the more complex the text is. Therefore, in order to characterize the complexity of the text, it is necessary to quantify the text length and text correlation.

[0072] In one embodiment, the step S300: determining the text length and text relevance of the preprocessed text according to the preprocessed text information includes:

[0073] S310, calling a text length function to perform text length identification on the preprocessed text information to obtain the text length of the preprocessed text;

[0074] S320: Input the preprocessed text into a deep learning model to estimate the text relevance to obtain the text relevance of the preprocessed text.

[0075] Specifically, a text length function can be called to perform text length identification on the preprocessed text information to obtain the text length of the preprocessed text, and a deep learning model can be used to estimate the text relevance of the preprocessed text. For example, a convolutional neural network can be used to perform semantic association feature identification to perform association estimation to obtain the text relevance of the preprocessed text. The convolutional neural network can be obtained by configuring parameters after testing a large amount of data.

[0076] S400, determining a redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text, wherein the redundancy coefficient is used to characterize the complexity of the preprocessed text;

[0077] It can be understood that both text length and text relevance have a certain impact on the complexity of the text. Therefore, it is necessary to quantify the complexity of the text so as to accurately select the target data processing module according to the complexity of the text.

[0078] In one embodiment, the step S400: determining the redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text, comprises:

[0079] S410, inputting the text length and text relevance of the preprocessed text into a redundancy prediction model to obtain a redundancy coefficient value, wherein the redundancy prediction model satisfies the following expression:

[0080] C=(L / L0+1)*Ф1+(G / G0+1)*Ф2 (G≥0.5,L≤20 bytes)

[0081] C=(L / L0+1)*Ф2+(G / G0+1)*Ф1 (G<0.5,L>20 bytes)

[0082] In the formula, L0 is the length of the benchmark text, G0 is the relevance of the benchmark text, L is the text length of the preprocessed text, G is the text relevance of the preprocessed text, Ф1 and Ф2 are the corresponding influence weights, and Ф2>Ф1.

[0083] In an embodiment of the present application, the inventor obtains a benchmark text length L0 and a benchmark text relevance G0 corresponding to the communication server by optimizing the performance of the communication server. The following explanation is given using a benchmark text length of 120 bytes and a benchmark text relevance of 0.8 as an example. The text redundancy coefficients under the benchmark text length L0 and the benchmark text relevance G0 are both defined as 1.

[0084] Specifically, when constructing the model, the inventors separately evaluated the impact of text length on text complexity, and obtained multiple sets of experimental data on text length and server processing time: (5 bytes, 0.01s), (10 bytes, 0.019s), (15 bytes, 0.031s), (20 bytes, 0.038s), (25 bytes, 0.05s), (30 bytes, 0.058s), (35 bytes, 0.071s), (40 bytes, 0.082s), (45 bytes, 0.09s), (50 bytes, 0.105s), (55 bytes, 0.11s), (60 bytes, 0.121s), (70 bytes, 0.139s), (80 bytes, 0.16s), (90 bytes, 0.179s), (100 bytes, 0.21s), (110 bytes, 0.22s), (120 bytes, 0.242s). It can be seen from the experimental data that the larger the length of the preprocessed text, the longer the server takes to process the text, thus leading to the conclusion that the text complexity is greater. Based on a large amount of experimental data, it is found after data fitting that the text length is roughly linearly correlated with the text complexity. After optimizing the original data, the inventors obtain the linear relationship between the redundancy coefficient value and the text length, which is C=L / L0+1.

[0085] When constructing the model, the inventor separately evaluated the impact of text relevance on text complexity, and obtained multiple sets of experimental data on text relevance and server processing time: (0.1, 0.01s), (0.2, 0.021s), (0.3, 0.031s), (0.4, 0.039s), (0.5, 0.05s), (0.6, 0.058s), (0.7, 0.071s), (0.8, 0.081s). From the experimental data, it can be seen that the greater the text relevance of the preprocessed text, the longer the server takes to process the text, thus concluding that the greater the text complexity, and based on a large amount of experimental data, after data fitting, it was found that the text relevance and text complexity are roughly linearly correlated. After optimizing the original data, the inventor obtained the linear relationship between the redundancy coefficient value and the text relevance value, which is C=G / G0+1.

[0086] On the basis that the benchmark text length that the server can process is 120 bytes and the benchmark text relevance is 0.8, the inventors found in the tests under different text lengths and text relevance that when the text relevance is greater than or equal to 0.5 and the text length is less than 20 bytes, the text relevance has a greater impact on the text complexity. The weight of the text relevance on the text complexity is 59%, and the weight of the text length on the text complexity is 41%. When the text relevance is less than 0.5 and the text length is greater than 20 bytes, the text length has a greater impact on the text complexity. The weight of the text relevance on the text complexity is 40.2%, and the weight of the text length on the text complexity is 59.8%. To facilitate the representation of the model, the influence weights of 59% and 59.8% are unified (i.e., Ф2 in the model expression is approximately equal to 0.6), and the influence weights of 41% and 40.2% are unified (i.e., Ф1 in the model expression is approximately equal to 0.4); therefore, the above-mentioned redundant estimation model expression is obtained.

[0087] For example, when the call voice sent by the first terminal is to report the names of different customers, such as "Li XX, Hu XX, Wang XX, Song XX, Yu XX", etc., the text length of the translated preprocessed text is 30 bytes, and the text relevance is 0.2, the redundancy coefficient value C=1.25*0.6+1.25*0.4=1.25 is obtained by using the redundancy prediction model. In other examples, the corresponding redundancy coefficient value can be calculated according to the text length and text relevance, which will not be repeated here.

[0088] In other embodiments, if the text is too long and the text correlation is too large so that the server cannot process it, the system will remind the user to decompose and send the voice, for example, reduce the speech speed and make appropriate pauses.

[0089] S500, selecting a target data processing module according to the redundancy coefficient of the pre-processed text to perform text processing on the pre-processed text information to obtain target text information, wherein the target data processing module includes a lightweight data processing module and a high-level data processing module, and the complexity of the text processed by the high-level data processing module is greater than the complexity of the text processed by the lightweight data processing module;

[0090] After obtaining the redundancy coefficient value according to the redundancy prediction model, it is necessary to compare the current redundancy coefficient with the redundancy coefficient threshold to select a data processing module that meets the requirements for text data processing.

[0091] In one embodiment, the step S500: selecting a target data processing module to perform text processing on the preprocessed text information to obtain target text information according to the redundancy coefficient of the preprocessed text, comprises:

[0092] S510, determining that the redundancy coefficient of the pre-processed text is greater than or equal to the redundancy coefficient threshold, selecting a high-level data processing module to perform text processing on the pre-processed text information to obtain target text information, wherein the high-level data processing module corrects language element errors of the text during the text processing to obtain the target text information;

[0093] S520, determining that the redundancy coefficient of the preprocessed text is less than a redundancy coefficient threshold, selecting a lightweight data processing module to perform text processing on the preprocessed text information to obtain target text information, and the lightweight data processing module identifies and segments text pauses during the text processing to obtain target text information.

[0094] The inventor of the present application has evaluated the performance of the lightweight data processing module and the high-level data processing module, and has conducted multiple data training on simple and complex sentences, and obtained a redundancy coefficient threshold of 2.0. That is, during data training, texts with a redundancy coefficient lower than 2.0 can be quickly processed by the lightweight data processing module of the present application to obtain translated voice information, and the accuracy can also meet the requirements; texts with a redundancy coefficient higher than 2.0 can be accurately processed by the high-level data processing module of the present application to obtain accurately translated voice information. Therefore, the inventor of the present application uses a redundancy coefficient of 2.0 as the redundancy coefficient threshold when the benchmark text length is 120 bytes and the benchmark text relevance is 0.8.

[0095] Specifically, when the redundancy coefficient of the preprocessed text is greater than or equal to the redundancy coefficient threshold of 2.0, it means that the sentences during the call are relatively complex, and the lightweight data processing module cannot meet the requirements of translation accuracy, and the high-level data processing module can correct the language element errors of the text during the text processing process. Therefore, the high-level data processing module is selected to perform text processing on the preprocessed text information to obtain the target text information. When the redundancy coefficient of the preprocessed text is less than the redundancy coefficient threshold of 2.0, it means that the sentences during the call are relatively simple, and the lightweight data processing module can meet the requirements of translation accuracy. Only some simple processing is required for the text, such as optimizing and segmenting the text pauses. Since no error correction is required, the time for text detection and correction can be saved. Therefore, the lightweight data processing module is selected to perform text processing on the preprocessed text information to obtain the target text information.

[0096] In one embodiment, if the call voice sentence is complex and a first language element error is detected after the translation to obtain text information, a first language element error correction model can be used to correct the first language element error of the preprocessed text to obtain the target text information, and the first language element error correction model is an error correction model obtained by training an artificial intelligence machine based on user historical feedback data.

[0097] For example, the first language element error correction model is the "adverbial position correction" model. If the communication server receives the voice data "we don't need to go to school today If the storm is coming", it translates it into "we don't need to go to school today if the storm is coming", and it is detected that "we don't need to go to school today if the storm is coming" does not conform to Chinese grammar and there is an inaccurate translation. At this time, the first language element error correction model is trained through the user's historical feedback data to correct it and get "If the storm is coming, we don't need to go to school today" to obtain accurate target text information, thereby performing accurate voice translation.

[0098] In other embodiments, artificial intelligence machine training can be performed based on user historical feedback data to obtain a second language element error correction model. For example, the second language element error correction model is a "subject position correction" model. Similarly, when it is detected that the subject position of the translated text is wrong, the subject position can be corrected through the second language element error correction model to obtain accurate target text information, thereby performing accurate speech translation.

[0099] In one embodiment, if the conversation speech is simple but has obvious intervals, after obtaining the preprocessed text, the preprocessed text can be feature recognized according to the preset recognition rules to obtain the segmentation boundary; then the separator is inserted before or after the segmentation boundary to obtain the target text information. And the speech pause time interval can be determined according to the mapping table of the relationship between the separator type and time; then according to the speech pause time interval, a speech pause of a corresponding time is performed in the process of converting the text on both sides of the separator into speech.

[0100] For example, when the user reports his name through the first terminal, he can pause according to the "surname", such as making an appropriate pause before the "surname", such as Hu XX (interval 1S) Li XX (interval 1S) Wang XX. In this way, the target text information that meets the accuracy requirements is obtained.

[0101] S600, converting the target text information into speech audio data in a second target language;

[0102] After obtaining relatively accurate target text information that conforms to the grammar of the target language, the target text information can be synthesized and converted into speech audio data in a second target language according to speech synthesis technology.

[0103] S700: Send the voice audio data in the second target language to the second terminal according to preset voice transmission configuration information.

[0104] After the speech audio data in the second target language is obtained, the speech audio data in the second target language may be sent to the second terminal according to the speech sending configuration rule.

[0105] In one embodiment, the step S700: sending the speech audio data in the second target language to the second terminal according to the preset speech transmission configuration information includes:

[0106] S710, determining a target speech speed, a target sound volume, and a target voice character object according to the speech speed configuration information, the sound volume configuration information, and the voice character object configuration information;

[0107] S720: Send the speech audio data of the second target language to the second terminal according to one or more of the target speech speed, the target sound volume, and the target voice character object.

[0108] Specifically, voice audio data is generated according to the speaking speed, sound volume and voice character object configured by the terminal user. For example, when the second terminal user is an elderly person, the communication server can generate voice audio data of the second target language with a slower speaking speed and louder voice according to the configuration information, and send it to the second terminal; when the second terminal user has special needs, special voice character objects can also be configured to generate voice audio data, such as generating the voice of a certain film or television character, and sending it to the second terminal.

[0109] Based on this, the data processing method for call voice translation of the present application converts the voice translation of the first terminal into preprocessed text. When the redundancy coefficient of the preprocessed text is greater than or equal to the redundancy coefficient threshold, a high-level data processing module is selected to perform text processing on the preprocessed text information to obtain target text information; when the redundancy coefficient of the preprocessed text is less than the redundancy coefficient threshold, a lightweight data processing module is selected to perform text processing on the preprocessed text information to obtain target text information. In this way, different data processing modules are selected according to the complexity of the pre-processed text to avoid a single data processing unit performing data processing on all translation requests at the same time, resulting in insufficient computing power and reduced translation efficiency and accuracy. For pre-processed texts with low complexity, lightweight data processing modules are selected for data processing, without the need for error detection, correction and other procedures, thereby improving the efficiency of call voice translation; and for pre-processed texts with high complexity, high-level data processing modules are selected for data processing, and error detection, correction and other procedures are performed to improve the accuracy of call voice translation. Therefore, in the data processing method for call voice translation of the present application, the communication server can simultaneously perform targeted translation processing on the call voices of different terminals, thereby maximizing the function of the communication server and meeting the translation needs of different voice calls.

[0110] The present application also provides a data processing system for call speech translation. Figure 3 , Figure 3 A hardware structure diagram of a data processing system for call speech translation provided in some embodiments of the present application, wherein the data processing system for call speech translation includes a memory 110 and a processor 120, wherein the memory 110 is used to store program code, and the processor 120 is used to call the program code to execute the method described above.

[0111] The processor 120 is used to provide computing and control capabilities to control the data processing system of the call voice translation to perform corresponding tasks, for example, to control the data processing system of the call voice translation to perform the data processing method of the call voice translation in any of the above-mentioned method embodiments, the method comprising: obtaining voice audio data sent by the first terminal, wherein the voice audio data is voice audio data in a first target language; according to preset translation configuration information, translating the voice audio data sent by the first terminal to obtain pre-processed text information in a second target language; determining the text length and text relevance of the pre-processed text according to the pre-processed text information; and determining the text length and text relevance of the pre-processed text according to the text length and text relevance of the pre-processed text. This correlation determines the redundancy coefficient of the preprocessed text, and the redundancy coefficient is used to characterize the complexity of the preprocessed text; according to the redundancy coefficient of the preprocessed text, a target data processing module is selected to perform text processing on the preprocessed text information to obtain target text information, wherein the target data processing module includes a lightweight data processing module and a high-level data processing module, and the complexity of the text processed by the high-level data processing module is greater than the complexity of the text processed by the lightweight data processing module; the target text information is converted into speech and audio data in a second target language; and the speech and audio data in the second target language are sent to a second terminal according to preset voice transmission configuration information.

[0112] The processor 120 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a hardware chip, or any combination thereof; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The above-mentioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0113] The memory 110 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the data processing method for call speech translation in the embodiment of the present application. The processor 120 can implement the data processing method for call speech translation in any of the above method embodiments by running the non-transitory software programs, instructions and modules stored in the memory 110.

[0114] Specifically, the memory 110 may include a volatile memory (VM), such as a random access memory (RAM); the memory 110 may also include a non-volatile memory (NVM), such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD) or other non-transitory solid-state storage device; the memory 110 may also include a combination of the above types of memory.

[0115] In summary, the data processing system for call speech translation of the present application adopts the technical solution of any one of the above-mentioned data processing method embodiments for call speech translation, and therefore, has at least the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be described one by one here.

[0116] The embodiment of the present application also provides a computer-readable storage medium, such as a memory including a program code, and the program code can be executed by a processor to complete the data processing method of the call speech translation in the above embodiment. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CDROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0117] The present application also provides a computer program product, which includes one or more program codes stored in a computer-readable storage medium. The processor of the data processing system for call speech translation reads the program code from the computer-readable storage medium, and the processor executes the program code to complete the steps of the data processing method for call speech translation provided in the above embodiment.

[0118] A person skilled in the art will appreciate that all or part of the steps for implementing the above embodiments may be accomplished by hardware or by hardware associated with a program code, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0119] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0120] Through the description of the above embodiments, it can be clearly understood by those skilled in the art that each embodiment can be implemented by means of software plus a general hardware platform, or by hardware. It can be understood by those skilled in the art that all or part of the processes in the above embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0121] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A data processing method for call speech translation, characterized in that: The method comprises: Acquire voice and audio data sent by the first terminal, wherein the voice and audio data is voice and audio data in a first target language; According to preset translation configuration information, the voice audio data sent by the first terminal is translated to obtain preprocessed text information in a second target language; Determining the text length and text relevance of the preprocessed text according to the preprocessed text information; Determining a redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text, wherein the redundancy coefficient is used to characterize the complexity of the preprocessed text; According to the redundancy coefficient of the pre-processed text, a target data processing module is selected to perform text processing on the pre-processed text information to obtain target text information, wherein the target data processing module includes a lightweight data processing module and a high-level data processing module, and the complexity of the text processed by the high-level data processing module is greater than the complexity of the text processed by the lightweight data processing module; Converting the target text information into speech audio data in a second target language; sending the voice audio data in the second target language to the second terminal according to preset voice transmission configuration information; The step of selecting a target data processing module according to the redundant coefficient of the preprocessed text to perform text processing on the preprocessed text information to obtain target text information includes: Determine that the redundancy coefficient of the preprocessed text is greater than or equal to the redundancy coefficient threshold, select a high-level data processing module to perform text processing on the preprocessed text information to obtain target text information, and the high-level data processing module corrects language element errors in the text during the text processing to obtain the target text information; Determine that the redundancy coefficient of the preprocessed text is less than a redundancy coefficient threshold, select a lightweight data processing module to perform text processing on the preprocessed text information to obtain target text information, and the lightweight data processing module identifies and segments text pauses during the text processing to obtain target text information.

2. The data processing method for call speech translation according to claim 1, characterized in that: The translation configuration information includes language configuration information, and the step of translating the voice audio data sent by the first terminal to obtain preprocessed text information in a second target language according to the preset translation configuration information includes: According to preset language configuration information, the voice audio data sent by the first terminal is translated to obtain preprocessed text information in a second target language, wherein the language configuration information is used to represent the translation of the first target language into the second target language, or the translation of the second target language into the first target language.

3. The data processing method for call speech translation according to claim 1, characterized in that: Before determining the text length and text relevance of the preprocessed text according to the preprocessed text information, the method further includes: detecting whether there is text noise in the preprocessed text information in the second target language; If text noise exists in the preprocessed text information in the second target language, the text noise in the preprocessed text information in the second target language is filtered.

4. The data processing method for call speech translation according to claim 1, characterized in that: The step of determining the text length and text relevance of the preprocessed text according to the preprocessed text information includes: Calling a text length function to perform text length identification on the preprocessed text information to obtain the text length of the preprocessed text; The preprocessed text is input into a deep learning model to estimate the text relevance to obtain the text relevance of the preprocessed text.

5. The data processing method for call speech translation according to claim 1, characterized in that: The determining the redundancy coefficient of the preprocessed text according to the text length and text relevance of the preprocessed text comprises: The text length and text relevance of the preprocessed text are input into a redundant prediction model to obtain a redundant coefficient value, wherein the redundant prediction model satisfies the following expression: C=(L / L0+1)*Ф1+(G / G0+1)*Ф2(G≥0.5, L≤20 bytes) C = (L / L0+1)*Ф2+(G / G0+1)*Ф1 (G<0.5, L>20 bytes) In the formula, L0 is the length of the benchmark text, G0 is the relevance of the benchmark text, L is the text length of the preprocessed text, G is the text relevance of the preprocessed text, Ф1 and Ф2 are the corresponding influence weights, and Ф2>Ф1.

6. The data processing method for call speech translation according to claim 1, characterized in that: The high-level data processing module includes a first language element error correction model, and the high-level data processing module is selected to perform text processing on the pre-processed text information to obtain target text information, including: detecting whether the preprocessed text has a first language element error; It is determined that a first language element error exists in the preprocessed text, and the first language element error correction model is used to correct the first language element error in the preprocessed text to obtain target text information, wherein the first language element error correction model is an error correction model obtained by artificial intelligence machine training based on user historical feedback data.

7. The data processing method for call speech translation according to claim 1, characterized in that: The high-level data processing module includes a second language element error correction model, and the high-level data processing module is selected to perform text processing on the pre-processed text information to obtain target text information, including: detecting whether the preprocessed text has a second language element error; It is determined that there are second language element errors in the preprocessed text, and the second language element error correction model is used to correct the second language element errors in the preprocessed text to obtain target text information, wherein the second language element error correction model is an error correction model obtained by artificial intelligence machine training based on user historical feedback data.

8. The data processing method for call speech translation according to claim 1, characterized in that: The lightweight data processing module includes a separator insertion algorithm, and the lightweight data processing module is selected to perform text processing on the pre-processed text information to obtain target text information, including: According to a preset recognition rule, feature recognition is performed on the preprocessed text to obtain a segmentation boundary; A separator insertion operation is performed before or after the segmentation boundary to obtain target text information.

9. The data processing method for call speech translation according to claim 8, characterized in that: The step of converting the target text information into speech audio data in a second target language includes: Determine the speech pause time interval according to a mapping table of the relationship between the spacer type and the time; According to the voice pause time interval, a voice pause of a corresponding time is performed during the text conversion process on both sides of the separator.

10. The data processing method for call speech translation according to claim 1, characterized in that: The voice transmission configuration information includes one or more of speech speed configuration information, sound volume configuration information, and voice character object configuration information. The sending of the voice audio data in the second target language to the second terminal according to the preset voice transmission configuration information includes: Determine the target speech speed, target sound volume and target voice character object according to the speech speed configuration information, the sound volume configuration information and the voice character object configuration information; The speech audio data of the second target language is sent to the second terminal according to one or more of the target speech speed, the target sound volume and the target speech character object.

11. A data processing system for call speech translation, characterized in that: include: A memory, the memory being used to store program codes; as well as, A processor, wherein the processor is used to call the program code to execute the method according to any one of claims 1 to 10.

Images