Transliteration model training method and apparatus, device, and storage medium

Abstract

The disclosure provides a transliteration model training method and device, equipment and a storage medium, relating to the fields of deep learning, natural language processing and the like artificial intelligence, and the implementation scheme is as follows: a plurality of sample texts to be translated of a first language type are obtained; a plurality of labeled sample texts corresponding to the plurality of sample texts to be translated are obtained, the first language type and a second language type are different, and the second language type is a language type corresponding to the labeled sample texts; a plurality of labeled pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of labeled sample texts is determined; and an initial transliteration model is trained according to the plurality of sample texts to be translated, the plurality of labeled sample texts, and the plurality of labeled pronunciation information to obtain a target transliteration model. Thus, the feature of the pronunciation dimension between the sample texts to be translated and the labeled sample texts is effectively modeled, the influence of pronunciation ambiguity on the translation effect is avoided, and the translation accuracy and effect are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, in particular to the field of artificial intelligence technologies such as deep learning and natural language processing, and specifically to a training method, apparatus, device, and storage medium for a transliteration model. Background Art

[0002] Artificial intelligence (AI) is the study of how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily encompass computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graphs.

[0003] In related technologies, rule-based matching methods are usually adopted, or sequence labeling is performed on massive training data to assist in converting input text of language type A into output text of language type B in practical applications. In this way, the translation may be inaccurate, inefficient, and have poor results due to pronunciation ambiguity. Summary of the Invention

[0004] The present disclosure provides a training method, apparatus, device, and storage medium for a transliteration model.

[0005] According to a first aspect of the present disclosure, a method for training a transliteration model is provided, comprising: obtaining a plurality of sample texts to be translated of a first language type; obtaining a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, wherein the first language type and a second language type are different, and the second language type is the language type corresponding to the annotated sample texts; determining a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts; and training an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model.

[0006] According to a second aspect of the present disclosure, a transliteration method is provided, comprising: obtaining a text to be translated in a first language type; and inputting the text to be translated into a target transliteration model trained using the above-mentioned transliteration model training method to obtain a predicted text output by the target transliteration model, wherein the first language type and a second language type are different, and the second language type is a language type corresponding to the predicted text.

[0007] According to a third aspect of the present disclosure, a transliteration model training device is provided, comprising: a first acquisition module for acquiring a plurality of sample texts to be translated of a first language type; a second acquisition module for acquiring a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts; a determination module for determining a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts; and a training module for training an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model.

[0008] According to a fourth aspect of the present disclosure, a transliteration device is provided, comprising: a third acquisition module for acquiring a text to be translated in a first language type; and a fourth acquisition module for inputting the text to be translated into a target transliteration model trained by the transliteration model training device, so as to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text.

[0009] According to the fifth aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable 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 training method of the transliteration model as described in the first aspect, or execute the transliteration method as described in the second aspect.

[0010] According to a sixth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the training method of the transliteration model as described in the first aspect, or to execute the transliteration method as described in the second aspect.

[0011] According to a seventh aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the training method for the transliteration model as described in the first aspect, or implements the transliteration method as described in the second aspect.

[0012] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0015] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;

[0016] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure;

[0017] Figure 4 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0018] Figure 5 is a schematic diagram of application effects in an embodiment of the present disclosure;

[0019] Figure 6 is a schematic diagram of a transliteration method in an embodiment of the present disclosure;

[0020] Figure 7 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0021] Figure 8 is a schematic diagram according to a sixth embodiment of the present disclosure;

[0022] Figure 9 is a schematic diagram according to a seventh embodiment of the present disclosure;

[0023] Figure 10 is a block diagram of an electronic device for implementing a training method of a transliteration model in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding, and should be considered as merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, in order to be clear and concise, descriptions of well-known functions and structures are omitted in the following description.

[0025] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure.

[0026] It should be noted that the execution subject of the training method of the transliteration model in the present embodiment is a training device of the transliteration model, which can be implemented in software and / or hardware, and can be configured in an electronic device, which can include but is not limited to a terminal, a server, etc.

[0027] Embodiments of the present disclosure relate to the technical field of artificial intelligence such as deep learning and natural language processing.

[0028] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0029] Deep learning involves learning the inherent patterns and representational hierarchies of sample data. The information gained from this learning process is highly helpful in interpreting data such as text, images, and sounds. The ultimate goal of deep learning is to enable machines to have the same analytical and learning capabilities as humans, enabling them to recognize data such as text, images, and sounds.

[0030] Natural language processing is a variety of theories and methods that can enable effective communication between humans and computers using natural language.

[0031] In the embodiments of the present disclosure, a method for training a translation model is provided to solve the technical problems in related technologies that may lead to inaccurate translation, low translation efficiency, and poor translation results due to pronunciation ambiguity. Since the transliteration model is trained by referring to multiple annotated pronunciation information between multiple sample texts to be translated and corresponding multiple annotated sample texts, it can effectively model the characteristics of the pronunciation dimension between the sample texts to be translated and the annotated sample texts, avoiding the influence of pronunciation ambiguity on the translation effect, thereby improving the translation accuracy and translation effect.

[0032] like Figure 1 As shown, the training method of the transliteration model includes:

[0033] S101: Acquire a plurality of sample texts to be translated in a first language.

[0034] Among them, the first language type can be, for example, English, Korean, Japanese, etc., the sample text to be translated can be a sample text to be translated, such as a paragraph of text in the first language type, etc., and the sample text can be a text used to train a transliteration model, and there is no restriction on this.

[0035] Correspondingly, the second language type described below may also be English, Korean, Japanese, etc., while the first language type and the second language type are different. Thus, the transliteration model obtained by the training implemented in the present disclosure supports translating the sample text to be translated in the first language type into the output text in the second language type. The output text of the transliteration model may be referred to as the predicted text described below, without limitation.

[0036] In the embodiment of the present disclosure, an example is given in which the first language type is Latin and the second language type is Indian, but this is not limited to this.

[0037] The plurality of sample texts to be translated in the first language type mentioned above may be a plurality of sample texts to be translated in Latin, such as “kripya”, which is not limited thereto.

[0038] S102: Acquire a plurality of annotated sample texts respectively corresponding to a plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts.

[0039] After obtaining the plurality of sample texts to be translated in the first language type as described above, a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated can be obtained. Among them, the sample texts used for annotation for training the transliteration model can be referred to as annotated sample texts. The annotated sample texts can be used as reference annotations when training the transliteration model. For details, please refer to the subsequent embodiments.

[0040] The multiple annotated sample texts corresponding to the multiple sample texts to be translated may specifically be sample texts of the second language type whose pronunciation features are similar to those of the sample texts to be translated.

[0041] For example, assuming that the sample text to be translated is “kripya”, then the corresponding multiple annotated sample texts may be Among them The language type is Indian. The pronunciation characteristics of the sample text to be translated, such as "kripya", are similar to The pronunciation characteristics are similar.

[0042] The pronunciation features of the sample text to be translated can be the corresponding tone, phoneme, vowel, and consonant features when the voice output is "kripya", and the pronunciation features of the annotated sample text can be the corresponding tone, phoneme, vowel, and consonant features when the voice output is the annotated sample text, and there is no restriction on this.

[0043] That is, the embodiments of the present disclosure support using a plurality of sample texts to be translated having similar pronunciation features and the pronunciation features between the corresponding plurality of annotated sample texts to assist in training the transliteration model.

[0044] S103: Determine a plurality of annotated pronunciation information between a plurality of sample texts to be translated and a plurality of corresponding annotated sample texts.

[0045] After obtaining multiple sample texts to be translated and corresponding multiple annotated sample texts, the pronunciation features corresponding to the multiple sample texts to be translated and the pronunciation features corresponding to the multiple annotated sample texts can be analyzed, thereby learning the similarity information between the pronunciation features corresponding to the sample texts of the two language types, and then using the similarity information as the above-mentioned annotated pronunciation information. Alternatively, the feature correlation between the pronunciation features corresponding to the sample texts of the two language types can be learned, and the feature correlation can be used as the above-mentioned annotated pronunciation information. There is no limitation to this.

[0046] The sample text to be translated may have corresponding annotated pronunciation information for each annotated sample text, so that multiple annotated pronunciation information may correspond to multiple sample texts to be translated.

[0047] S104: training an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model.

[0048] After the multiple annotated pronunciation information between the multiple sample texts to be translated and the corresponding multiple annotated sample texts, the initial transliteration model can be trained based on the multiple sample texts to be translated, the multiple annotated sample texts, and the multiple annotated pronunciation information to obtain a target transliteration model, thereby realizing the transliteration model trained by referring to the multiple annotated pronunciation information between the multiple sample texts to be translated and the corresponding multiple annotated sample texts (the trained transliteration model can be referred to as the target transliteration model).

[0049] The above-mentioned initial transliteration model can be an artificial intelligence model, such as a neural network model, a machine learning model, or a graph neural network model. Of course, any other possible model that can perform the translation task can also be used, for example, a Hidden Markov Model (HMM), without limitation.

[0050] That is to say, multiple sample texts to be translated, multiple annotated sample texts, and multiple annotated pronunciation information can be input into the initial transliteration model until the transliteration model meets certain convergence conditions, and the trained transliteration model is used as the target transliteration model.

[0051] Optionally, in some embodiments, multiple sample texts to be translated can be input into an initial transliteration model to obtain multiple predicted sample texts output by the transliteration model, and multiple predicted pronunciation information between the multiple sample texts to be translated and the corresponding multiple predicted sample texts can be determined. If the multiple predicted pronunciation information and the multiple annotated pronunciation information meet the set conditions, the transliteration model training is completed, thereby realizing the determination of the convergence timing of the model based on the predicted pronunciation information and the annotated pronunciation information, so that the trained target transliteration model can have a more accurate prediction representation capability for the pronunciation characteristics of the text, effectively improving the prediction accuracy of the model.

[0052] Among them, multiple sample texts to be translated are input into the initial transliteration model to obtain the sample text output by the transliteration model, which can be called predicted sample text. The transliteration model can analyze the pronunciation features corresponding to the sample texts to be translated, and then identify and parse the sample texts of the second language type that are similar to the pronunciation features and use them as predicted sample texts.

[0053] The pronunciation features of the above-mentioned predicted sample text can be called predicted pronunciation information. The predicted pronunciation information can be the corresponding tone, phoneme, vowel, and consonant features when the predicted sample text is output by voice, and there is no limitation on this.

[0054] The disclosed embodiment also supports pre-configuring a loss function for the initial transliteration model, using the loss function to fit the loss value between multiple predicted pronunciation information and multiple annotated pronunciation information, and pre-configuring a loss threshold. If the loss value between the multiple predicted pronunciation information and the multiple annotated pronunciation information is less than the loss threshold, it can be determined that the multiple predicted pronunciation information and the multiple annotated pronunciation information meet the set conditions, and the transliteration model training is completed. Otherwise, the transliteration model is continuously trained dynamically using multiple sample texts to be translated, and there is no restriction on this.

[0055] In this embodiment, by obtaining a plurality of sample texts to be translated of a first language type, obtaining a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample text, and determining a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts, and training an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model. Since the transliteration model is trained with reference to the plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts, it is possible to effectively model the features of the pronunciation dimension between the sample texts to be translated and the annotated sample texts, thereby avoiding the influence of pronunciation ambiguity on the translation effect, thereby improving the translation accuracy and translation effect.

[0056] Figure 2is a schematic diagram according to the second embodiment of the present disclosure.

[0057] like Figure 2 As shown, the training method of the transliteration model includes:

[0058] S201: Acquire a plurality of sample texts to be translated in a first language.

[0059] S202: Acquire a plurality of annotated sample texts corresponding to a plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts.

[0060] The description of S201 - S202 can be specifically referred to the above embodiment and will not be repeated here.

[0061] S203: Determine a plurality of first pronunciation features corresponding to a plurality of sample texts to be translated.

[0062] S204: Determine a plurality of second pronunciation features corresponding to the plurality of annotated sample texts.

[0063] The pronunciation features of the sample text to be translated can be the corresponding tone, phoneme, vowel, and consonant features when the voice output is "kripya", and the pronunciation features of the annotated sample text can be the corresponding tone, phoneme, vowel, and consonant features when the voice output is the annotated sample text, and there is no restriction on this.

[0064] The pronunciation features of the sample text to be translated can be called first pronunciation features, and the pronunciation features of the annotated sample text can be called second pronunciation features.

[0065] S205: Determine a plurality of first pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of second pronunciation features, and use the plurality of pronunciation similarity information as a plurality of annotated pronunciation information.

[0066] After determining the multiple first pronunciation features corresponding to the multiple sample texts to be translated, and determining the multiple second pronunciation features corresponding to the multiple annotated sample texts, a similarity fitting function can be used to determine the similarity between the multiple first pronunciation features and the corresponding multiple second pronunciation features, and the similarity can be used as the first pronunciation similarity information, so that the first pronunciation similarity information can represent the degree of similarity between the multiple first pronunciation features and the corresponding multiple second pronunciation features. There is no limitation on this.

[0067] The above-mentioned method determines multiple first pronunciation features corresponding to multiple sample texts to be translated, and then determines multiple second pronunciation features corresponding to multiple annotated sample texts, so as to achieve the learning of first pronunciation similarity information between pronunciation features corresponding to sample texts of two language types, and then uses the first pronunciation similarity information as the above-mentioned annotated pronunciation information. Since the first pronunciation similarity information can represent the similarity between multiple first pronunciation features and the corresponding multiple second pronunciation features, the similarity between the pronunciation features can be quantified, so that the transliteration model can be assisted in training based on the dimension of the pronunciation features, thereby improving the transliteration model's representation and analysis capabilities for the pronunciation similarity of sample texts of different language types.

[0068] S206: Inputting a plurality of sample texts to be translated into an initial transliteration model to obtain a plurality of predicted sample texts output by the transliteration model.

[0069] The description of S205-S206 can be found in the above embodiment and will not be repeated here.

[0070] S207: Determine a plurality of third pronunciation features corresponding to the plurality of predicted sample texts.

[0071] Among them, multiple sample texts to be translated are input into the initial transliteration model to obtain the sample text output by the transliteration model, which can be called predicted sample text, and the pronunciation features of the predicted sample text (the pronunciation features of the predicted sample text can be called the third pronunciation features) can be the tone, phoneme, vowel, and consonant features corresponding to the predicted sample text output by the transliteration model, and there is no restriction on this.

[0072] S208: Determine a plurality of second pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features, and use the plurality of second pronunciation similarity information as a plurality of predicted pronunciation information.

[0073] After determining the plurality of third pronunciation features corresponding to the plurality of predicted sample texts, a plurality of second pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features may be determined, and the plurality of second pronunciation similarity information may be used as a plurality of predicted pronunciation information.

[0074] The plurality of pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features may be referred to as a plurality of second pronunciation similarity information.

[0075] In this way, it is possible to learn the second pronunciation similarity information between the pronunciation features of the input sample text to be translated and the pronunciation features corresponding to the predicted sample text output by the model, and then use the second pronunciation similarity information as the above-mentioned predicted pronunciation information. The predicted pronunciation information can be used to timely determine the convergence timing of the model. Since the second pronunciation similarity information can characterize the similarity between the pronunciation features of the input multiple sample texts to be translated and the multiple third pronunciation features corresponding to the predicted sample text, it is possible to quantify the similarity between the pronunciation features of the input multiple sample texts to be translated and the multiple third pronunciation features corresponding to the predicted sample text, so as to assist in training the transliteration model based on the dimension of the pronunciation features and timely determine the convergence timing of the model. It not only improves the transliteration model's ability to represent and analyze the pronunciation similarity of sample texts of different language categories, but also can determine the convergence timing based on the pronunciation similarity obtained by analysis, so as to achieve both the training efficiency and training effect of the model.

[0076] S209: If the plurality of predicted pronunciation information and the plurality of annotated pronunciation information satisfy the set conditions, the transliteration model training is completed.

[0077] The description of S209 can be found in the above embodiment and will not be repeated here.

[0078] In this embodiment, because the transliteration model is trained by referring to multiple annotated pronunciation information between multiple sample texts to be translated and corresponding multiple annotated sample texts, it can effectively model the characteristics of the pronunciation dimension between the sample texts to be translated and the annotated sample texts, avoiding the impact of pronunciation ambiguity on translation results, thereby improving translation accuracy and translation results. Because the first pronunciation similarity information can represent the similarity between multiple first pronunciation features and corresponding multiple second pronunciation features, it can quantify the similarity between the pronunciation features, thereby assisting in the training of the transliteration model based on the pronunciation feature dimension, improving the transliteration model's ability to represent and analyze the pronunciation similarity of sample texts in different language categories. Since the second pronunciation similarity information can represent the similarity between the pronunciation features of the input multiple sample texts to be translated and the multiple third pronunciation features corresponding to the predicted sample texts, it is possible to quantify the similarity between the pronunciation features of the input multiple sample texts to be translated and the multiple third pronunciation features corresponding to the predicted sample texts, thereby assisting in training the transliteration model based on the dimension of the pronunciation features and timely determining the convergence timing of the model. This not only improves the transliteration model's ability to represent and analyze the pronunciation similarity of sample texts in different language categories, but also can determine the convergence timing based on the pronunciation similarity obtained by analysis, thereby achieving a balance between the training efficiency and training effect of the model.

[0079] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure.

[0080] like Figure 3 As shown, the training method of the transliteration model includes:

[0081] S301: Segmenting a plurality of sample texts to be translated respectively to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated respectively.

[0082] This embodiment provides a method for processing training data, that is, before training the initial transliteration model, the pronunciation features between the obtained sample text to be translated and the annotated sample text are aligned to help ensure the accuracy of the transliteration model's mapping of the pronunciation features.

[0083] That is to say, multiple sample texts to be translated can be segmented separately to obtain multiple sample sub-texts corresponding to the multiple sample texts to be translated. For example, based on certain segmentation rules, each sample text to be translated can be segmented into sample sub-texts of a set length. For example, "kripya" can be segmented into "kri" and "pya". Alternatively, any other possible method can be used to segment "kripya", and there is no limitation on this.

[0084] Optionally, in some embodiments, multiple sample texts to be translated can be segmented based on the syllable dimension to obtain multiple sample sub-texts corresponding to the multiple sample texts to be translated. The pronunciation of the sample sub-text corresponds to a pronunciation unit, thereby effectively improving the mapping accuracy of the transliteration model to the pronunciation features and significantly improving the prediction effect of the model.

[0085] For example, assuming that the multiple sample texts to be translated are "thoriyam", "dhaatu", and "vaayu", then the multiple sample texts to be translated are segmented based on the syllable dimension to obtain multiple sample sub-texts corresponding to the multiple sample texts to be translated. The multiple sample sub-texts can be "tho", "ri", "yam", "dhaa", "tu", "vaa", and "yu", where the pronunciation of each sample sub-text corresponds to a pronunciation unit, and a pronunciation unit can be composed of consonants and vowels, thereby ensuring the accuracy of pronunciation feature mapping while reducing the amount of training data.

[0086] S302: Determine a plurality of pronunciation sub-features corresponding to a plurality of sample sub-texts, and use the plurality of pronunciation sub-features as a plurality of first pronunciation features.

[0087] The above splitting the plurality of to-be-translated sample texts respectively to obtain a plurality of sample subtexts corresponding to the plurality of to-be-translated sample texts can determine a plurality of pronunciation sub-features corresponding to the plurality of sample subtexts, and take the plurality of pronunciation sub-features as a plurality of first pronunciation features. Since the pronunciation sub-features obtained by splitting are used to assist in training the phonetic translation model, the workload of later labeling can be effectively reduced, thereby reducing the size of the selected initial phonetic translation model, facilitating the deployment and implementation of the target phonetic translation model, and expanding the deployment and installation scenarios of the target phonetic translation model.

[0088] S303: Aligning the plurality of pronunciation sub-features and the corresponding plurality of pronunciation features.

[0089] The pronunciation feature is a second pronunciation feature or a third pronunciation feature.

[0090] That is, after splitting the plurality of pronunciation sub-features, the plurality of pronunciation sub-features and the corresponding plurality of second pronunciation features can be aligned, and the second pronunciation feature is the pronunciation feature corresponding to the labeled sample text, so as to quickly and accurately model the phonetic similarity information between the pronunciation feature of the to-be-translated sample text and the pronunciation feature corresponding to the labeled sample text by the phonetic translation model.

[0091] In addition, after splitting the plurality of pronunciation sub-features, the plurality of pronunciation sub-features and the corresponding plurality of third pronunciation features can be aligned, and the third pronunciation feature is the pronunciation feature corresponding to the predicted sample text, so as to quickly and accurately model the phonetic similarity information between the pronunciation feature of the predicted sample text and the pronunciation feature of the to-be-translated sample text by the phonetic translation model, and guarantee the accuracy and representation efficiency of the mapping relationship representation between the pronunciation feature of the predicted sample text and the pronunciation feature of the to-be-translated sample text.

[0092] The above aligning the plurality of pronunciation sub-features and the corresponding plurality of second pronunciation features can be aligned by using a pre-trained alignment processing model, which can be, for example, a hidden Markov model (HMM), without limitation.

[0093] The above alignment process can be a word alignment process, that is, the sample subtext and the corresponding labeled sample text can be subjected to word alignment processing, so that after word alignment processing, the pronunciation sub-feature of the sample subtext and the second pronunciation feature of the labeled sample text have a corresponding mapping relationship. The mapping relationship obtained after the alignment processing can be used to assist the phonetic translation model in matching and predicting the input and output texts based on the pronunciation dimension.

[0094] The description of aligning the plurality of pronunciation sub-features and the corresponding plurality of third pronunciation features can be similarly described, without limitation.

[0095] S304: After the alignment process, determine multiple pronunciation similarity information between the multiple first pronunciation features and the corresponding multiple pronunciation features, wherein the pronunciation feature is the second pronunciation feature or the third pronunciation feature, and the pronunciation similarity information is the first pronunciation similarity information or the second pronunciation similarity information.

[0096] That is to say, after the above segmentation to obtain multiple pronunciation sub-features, and after the multiple pronunciation sub-features are aligned with the corresponding multiple second pronunciation features, and the multiple pronunciation similarity information between the multiple first pronunciation features and the corresponding multiple second pronunciation features is determined, the transliteration model can quickly and accurately model the pronunciation similarity information between the pronunciation features of the sample text to be translated and the pronunciation features corresponding to the annotated sample text, thereby ensuring the accuracy and efficiency of the mapping relationship representation between the pronunciation features of the sample text to be translated and the pronunciation features corresponding to the annotated sample text.

[0097] In addition, after the above segmentation to obtain multiple pronunciation sub-features and alignment of the multiple pronunciation sub-features with the corresponding multiple third pronunciation features, multiple pronunciation similarity information between the multiple first pronunciation features and the corresponding multiple third pronunciation features is determined, so that the transliteration model can quickly and accurately model the pronunciation similarity information between the pronunciation features of the predicted sample text and the pronunciation features of the sample text to be translated, thereby ensuring the accuracy and representation efficiency of the mapping relationship representation between the pronunciation features of the predicted sample text and the pronunciation features of the sample text to be translated.

[0098] In this embodiment, by segmenting a plurality of sample texts to be translated to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated, a plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts can be determined, and the plurality of pronunciation sub-features can be used as a plurality of first pronunciation features. Since the transliteration model is assisted in training based on the plurality of pronunciation sub-features obtained by segmentation, the workload of later annotation can be effectively reduced, thereby reducing the scale of the selected initial transliteration model, facilitating the deployment and implementation of the target transliteration model, and expanding the deployment and installation scenario of the target transliteration model. By segmenting a plurality of sample texts to be translated based on the syllable dimension to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated, the pronunciation of the sample sub-text corresponds to a pronunciation unit, thereby effectively improving the accuracy of the transliteration model's mapping of pronunciation features, and significantly improving the model's prediction effect. It can also facilitate the transliteration model to quickly and accurately model the pronunciation similarity information between the pronunciation features of the predicted sample text (or the pronunciation features of the annotated sample text) and the pronunciation features of the sample text to be translated, ensuring the accuracy and efficiency of the mapping relationship representation between the pronunciation features of the predicted sample text (or the pronunciation features of the annotated sample text) and the pronunciation features of the sample text to be translated.

[0099] Figure 4 is a schematic diagram according to a fourth embodiment of the present disclosure.

[0100] like Figure 4 As shown, the transliteration method includes:

[0101] S401: Obtain a text to be translated in a first language.

[0102] The text currently to be translated may be referred to as a text to be translated, and the text to be translated may be a text input by a user received by the electronic device based on a preset input method.

[0103] like Figure 5 As shown, Figure 5 It is a schematic diagram of the application effect in the embodiment of the present disclosure. Assuming that the electronic device is based on a Latin input method based on a preset input method, it can receive the text to be translated 51 input by the user, and then call the target transliteration model trained by the above embodiment to determine at least one predicted text 52, and provide at least one predicted text 52 to the user.

[0104] S402: Inputting the text to be translated into the target transliteration model as described above to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text.

[0105] After obtaining the text to be translated in the first language type, the text to be translated is input into the target transliteration model as described above, so that the text to be translated can be syllable-segmented based on the target transliteration model to obtain multiple sub-texts to be translated, and the corresponding predicted text can be determined based on the pronunciation sub-features of each sub-text to be translated, without any limitation.

[0106] In some other embodiments, the at least one predictive text provided to the user may be selected from a plurality of predictive texts.

[0107] For example, if Figure 6 As shown, Figure 6 It is a flowchart of the transliteration method in the embodiment of the present disclosure. Assume that the text to be translated is divided into multiple sub-texts to be translated "ha", "rry", "r", and "ry". Then, at least one predicted text 61 can be obtained by using the target transliteration model. The probability value corresponding to each predicted text 61 can also be determined, so that the predicted text on the path with the larger sum of probability values ​​can be used as the predicted text displayed to the user, and there is no restriction on this.

[0108] The probability value mapped to each predicted text 61 may be determined by combining a pronunciation-related application, a Viterbi algorithm, and a search algorithm, and there is no limitation to this.

[0109] In this embodiment, by obtaining a text to be translated of a first language type, the text to be translated is input into the target transliteration model as described above to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text. Since the predicted text is assisted in determining based on the target transliteration model obtained through the above training, and the target transliteration model is trained with reference to multiple annotated pronunciation information between multiple sample texts to be translated and corresponding multiple annotated sample texts, the corresponding predicted text is translated based on the characteristics of the pronunciation dimension of the text to be translated, thereby avoiding the influence of pronunciation ambiguity on the translation effect, thereby improving the translation accuracy and translation effect.

[0110] Figure 7 is a schematic diagram according to a fifth embodiment of the present disclosure.

[0111] like Figure 7 As shown, the transliteration model training device 70 includes:

[0112] A first acquisition module 701 is used to acquire a plurality of sample texts to be translated in a first language type;

[0113] A second acquisition module 702 is configured to acquire a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts;

[0114] A determination module 703 is configured to determine a plurality of annotated pronunciation information between a plurality of sample texts to be translated and a plurality of corresponding annotated sample texts; and

[0115] The training module 704 is configured to train an initial transliteration model based on a plurality of sample texts to be translated, a plurality of annotated sample texts, and a plurality of annotated pronunciation information to obtain a target transliteration model.

[0116] In some embodiments of the present disclosure, Figure 8 As shown, Figure 8 is a schematic diagram according to the sixth embodiment of the present disclosure, wherein the transliteration model training device 80 includes: a first acquisition module 801, a second acquisition module 802, a determination module 803, and a training module 804, wherein the training module 804 includes:

[0117] The acquisition submodule 8041 is used to input a plurality of sample texts to be translated into the initial transliteration model to obtain a plurality of predicted sample texts output by the transliteration model;

[0118] The first determining submodule 8042 is configured to determine a plurality of predicted pronunciation information between a plurality of sample texts to be translated and a plurality of corresponding predicted sample texts;

[0119] The training submodule 8043 is configured to complete the transliteration model training when a set condition is satisfied between a plurality of predicted pronunciation information and a plurality of annotated pronunciation information.

[0120] In some embodiments of the present disclosure, Figure 8 As shown, the determination module 803 includes:

[0121] The second determining submodule 8031 ​​is used to determine a plurality of first pronunciation features corresponding to a plurality of sample texts to be translated;

[0122] The third determining submodule 8032 is configured to determine a plurality of second pronunciation features corresponding to the plurality of annotated sample texts;

[0123] The fourth determining submodule 8033 is configured to determine a plurality of first pronunciation similarity information between a plurality of first pronunciation features and a plurality of corresponding second pronunciation features, and use the plurality of pronunciation similarity information as a plurality of annotated pronunciation information.

[0124] In some embodiments of the present disclosure, the first determining submodule 8042 is specifically configured to:

[0125] Determining a plurality of first pronunciation features corresponding to a plurality of sample texts to be translated;

[0126] Determining a plurality of third pronunciation features corresponding to the plurality of predicted sample texts respectively;

[0127] A plurality of second pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features is determined, and the plurality of second pronunciation similarity information is used as a plurality of predicted pronunciation information.

[0128] In some embodiments of the present disclosure, the second determining submodule 8031 ​​is specifically configured to:

[0129] Segmenting the plurality of sample texts to be translated respectively to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated respectively;

[0130] A plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts are determined, and the plurality of pronunciation sub-features are used as a plurality of first pronunciation features.

[0131] In some embodiments of the present disclosure, the second determining submodule 8031 ​​is specifically configured to:

[0132] The plurality of sample texts to be translated are segmented based on the syllable dimension to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated, wherein the pronunciation of the sample sub-text corresponds to a pronunciation unit.

[0133] In some embodiments of the present disclosure, the second determining submodule 8031 ​​is specifically configured to:

[0134] After determining a plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts, aligning the plurality of pronunciation sub-features with the corresponding plurality of pronunciation features;

[0135] After the alignment process, determining a plurality of pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of pronunciation features;

[0136] The pronunciation feature is the second pronunciation feature or the third pronunciation feature, and the pronunciation similarity information is the first pronunciation similarity information or the second pronunciation similarity information.

[0137] It is understandable that the present embodiment Figure 8 The training device 80 of the transliteration model and the training device 70 of the transliteration model in the above embodiment, the first acquisition module 801 and the first acquisition module 701 in the above embodiment, the second acquisition module 802 and the second acquisition module 702 in the above embodiment, the determination module 803 and the determination module 703 in the above embodiment, and the training module 804 and the training module 704 in the above embodiment may have the same function and structure.

[0138] It should be noted that the aforementioned explanation of the transliteration model training method is also applicable to the transliteration model training device of this embodiment and will not be repeated here.

[0139] In this embodiment, by obtaining a plurality of sample texts to be translated of a first language type, obtaining a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample text, and determining a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts, and training an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model. Since the transliteration model is trained with reference to the plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts, it is possible to effectively model the features of the pronunciation dimension between the sample texts to be translated and the annotated sample texts, thereby avoiding the influence of pronunciation ambiguity on the translation effect, thereby improving the translation accuracy and translation effect.

[0140] Figure 9 is a schematic diagram according to a seventh embodiment of the present disclosure.

[0141] like Figure 9 As shown, the transliteration device 90 includes:

[0142] The third acquisition module 901 is configured to acquire a text to be translated in a first language.

[0143] The fourth acquisition module 902 is used to input the text to be translated into the target transliteration model trained by the transliteration model training device as described above, to obtain the predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text.

[0144] It should be noted that the above explanation of the transliteration method is also applicable to the transliteration device of this embodiment and will not be repeated here.

[0145] In this embodiment, by obtaining a text to be translated of a first language type, the text to be translated is input into the target transliteration model as described above to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text. Since the predicted text is assisted in determining based on the target transliteration model obtained through the above training, and the target transliteration model is trained with reference to multiple annotated pronunciation information between multiple sample texts to be translated and corresponding multiple annotated sample texts, the corresponding predicted text is translated based on the characteristics of the pronunciation dimension of the text to be translated, thereby avoiding the influence of pronunciation ambiguity on the translation effect, thereby improving the translation accuracy and translation effect.

[0146] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0147] Figure 10 is a block diagram of an electronic device for implementing the training method of the transliteration model of an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0148] like Figure 10As shown, the device 1000 includes a computing unit 1001 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded into a random access memory (RAM) 1003 from a storage unit 1008. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0149] A plurality of components in the device 1000 are connected to the I / O interface 1005, including: an input unit 1006, such as a keyboard, a mouse, and the like; an output unit 1007, such as various types of displays, speakers, and the like; a storage unit 1008, such as a magnetic disk, an optical disk, and the like; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0150] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, and the like. The computing unit 1001 performs various methods and processes described above, such as the training method of the transliteration model, or the transliteration method.

[0151] For example, in some embodiments, the training method of the transliteration model, or the transliteration method can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the training method of the transliteration model, or the transliteration method described above can be performed. Alternatively, in other embodiments, the computing unit 1001 can be configured to perform the training method of the transliteration model, or the transliteration method by any other appropriate means, such as by means of firmware.

[0152] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The program code for implementing the transliteration model training method or transliteration method disclosed herein can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program code is executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, 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 foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0156] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0157] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. This client-server relationship is established by computer programs running on the respective computers, establishing a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosts and VPS services ("Virtual Private Servers" or simply "VPS"). The server may also be a server in a distributed system or a server integrated with blockchain.

[0158] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this application can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.

[0159] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A transliteration model training method, comprising: Obtaining multiple sample texts to be translated in the first language; Acquire a plurality of annotated sample texts respectively corresponding to the plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts; Determining a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts; as well as Inputting the plurality of sample texts to be translated into an initial transliteration model to obtain a plurality of predicted sample texts output by the transliteration model, wherein the transliteration model analyzes pronunciation features corresponding to the sample texts to be translated, identifies and parses sample texts of the second language type that are similar to the pronunciation features and uses them as the predicted sample texts; Determining a plurality of predicted pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of predicted sample texts; If the plurality of predicted pronunciation information and the plurality of annotated pronunciation information satisfy a set condition, the transliteration model training is completed.

2. The method according to claim 1, wherein The determining of a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts includes: Determining a plurality of first pronunciation features respectively corresponding to the plurality of sample texts to be translated; Determining a plurality of second pronunciation features corresponding to the plurality of annotated sample texts respectively; A plurality of first pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of second pronunciation features is determined, and the plurality of pronunciation similarity information is used as the plurality of annotated pronunciation information.

3. The method according to claim 1, wherein The determining of a plurality of predicted pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of predicted sample texts includes: Determining a plurality of first pronunciation features corresponding to the plurality of sample texts to be translated; Determining a plurality of third pronunciation features respectively corresponding to the plurality of predicted sample texts; A plurality of second pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features is determined, and the plurality of second pronunciation similarity information is used as the plurality of predicted pronunciation information.

4. The method according to claim 2 or 3, wherein: The determining of a plurality of first pronunciation features respectively corresponding to the plurality of sample texts to be translated includes: Segmenting the plurality of sample texts to be translated respectively to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated respectively; A plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts are determined, and the plurality of pronunciation sub-features are used as the plurality of first pronunciation features.

5. The method according to claim 4, wherein The segmenting of the plurality of sample texts to be translated to obtain a plurality of sample sub-texts respectively corresponding to the plurality of sample texts to be translated comprises: The plurality of sample texts to be translated are segmented based on the syllable dimension to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated, wherein pronunciation of the sample sub-texts corresponds to a pronunciation unit.

6. The method according to claim 4, further comprising, after determining the plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts, performing alignment processing on the plurality of pronunciation sub-features and the corresponding plurality of pronunciation features; After the alignment process, determining a plurality of pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of pronunciation features; The pronunciation feature is the second pronunciation feature or the third pronunciation feature, and the pronunciation similarity information is the first pronunciation similarity information or the second pronunciation similarity information.

7. A transliteration method comprising: Obtain the text to be translated in the first language type; The text to be translated is input into a target transliteration model trained by the transliteration model training method according to any one of claims 1 to 6 above to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text.

8. A transliteration model training device, comprising: A first acquisition module is used to acquire a plurality of sample texts to be translated in a first language type; A second acquisition module is configured to acquire a plurality of annotated sample texts corresponding to the plurality of sample texts to be translated, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the annotated sample texts; a determination module, configured to determine a plurality of annotated pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of annotated sample texts; as well as A training module, configured to train an initial transliteration model based on the plurality of sample texts to be translated, the plurality of annotated sample texts, and the plurality of annotated pronunciation information to obtain a target transliteration model; The training module includes: an acquisition submodule, configured to input the plurality of sample texts to be translated into the initial transliteration model to obtain a plurality of predicted sample texts output by the transliteration model, wherein the transliteration model analyzes pronunciation features corresponding to the sample texts to be translated, identifies and parses sample texts of the second language type that are similar to the pronunciation features and uses them as the predicted sample texts; A first determining submodule is configured to determine a plurality of predicted pronunciation information between the plurality of sample texts to be translated and the corresponding plurality of predicted sample texts; The training submodule is configured to complete the training of the transliteration model when a set condition is satisfied between the plurality of predicted pronunciation information and the plurality of annotated pronunciation information.

9. The device according to claim 8, wherein The determining module includes: A second determining submodule is configured to determine a plurality of first pronunciation features corresponding to the plurality of sample texts to be translated; A third determining submodule is configured to determine a plurality of second pronunciation features corresponding to the plurality of annotated sample texts respectively; The fourth determining submodule is configured to determine a plurality of first pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of second pronunciation features, and use the plurality of pronunciation similarity information as the plurality of annotated pronunciation information.

10. The device according to claim 8, wherein The first determining submodule is specifically configured to: Determining a plurality of first pronunciation features corresponding to the plurality of sample texts to be translated; Determining a plurality of third pronunciation features respectively corresponding to the plurality of predicted sample texts; A plurality of second pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of third pronunciation features is determined, and the plurality of second pronunciation similarity information is used as the plurality of predicted pronunciation information.

11. The device according to claim 9, wherein The second determining submodule is specifically configured to: Segmenting the plurality of sample texts to be translated respectively to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated respectively; A plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts are determined, and the plurality of pronunciation sub-features are used as the plurality of first pronunciation features.

12. The device according to claim 11, wherein The second determining submodule is specifically configured to: The plurality of sample texts to be translated are segmented based on the syllable dimension to obtain a plurality of sample sub-texts corresponding to the plurality of sample texts to be translated, wherein pronunciation of the sample sub-texts corresponds to a pronunciation unit.

13. The apparatus according to claim 11, wherein the second determining submodule is specifically configured to: After determining the plurality of pronunciation sub-features corresponding to the plurality of sample sub-texts, aligning the plurality of pronunciation sub-features with the corresponding plurality of pronunciation features; After the alignment process, determining a plurality of pronunciation similarity information between the plurality of first pronunciation features and the corresponding plurality of pronunciation features; in, The pronunciation feature is the second pronunciation feature or the third pronunciation feature, and the pronunciation similarity information is the first pronunciation similarity information or the second pronunciation similarity information.

14. A transliteration device, comprising: A third acquisition module is used to acquire the text to be translated in the first language type; a fourth acquisition module, configured to input the text to be translated into a target transliteration model trained by the transliteration model training device according to any one of claims 8 to 13, to obtain a predicted text output by the target transliteration model, wherein the first language type and the second language type are different, and the second language type is the language type corresponding to the predicted text.

15. An electronic device comprising: at least one processor; as well as 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 to enable the at least one processor to perform the method according to any one of claims 1 to 6, or the method according to claim 7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 6, or to execute the method according to claim 7.

17. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 6, or performs the method according to claim 7.

Images