A subtitle translation quality detection method, device, equipment and medium

Abstract

The invention discloses a subtitle translation quality detection method, device, equipment and medium. The method performs speech recognition and image recognition on a bilingual subtitle translation video, and processes the original translation in the speech recognition result and the image recognition result into a bilingual sentence pair according to the subtitle timeline, then performs machine translation on the speech recognition result to obtain a machine-translated text, and uses the machine-translated text as a reference translation to perform quality detection on the original translation to obtain a quality score. When the obtained quality score is lower than a score threshold, a sentence pair correction model is used to correct the bilingual sentence pairs lower than the score threshold until the obtained quality score is not lower than the score threshold, so as to complete the detection of the subtitle translation quality, improve the quality of the subtitle translation, and ensure the reliability of subsequent use as a corpus resource.

Description

Technical Field

[0001] The present invention relates to the technical field of subtitle translation processing, and in particular to a subtitle translation quality detection method, device, equipment and medium. Background Art

[0002] As an important cultural carrier, TV movies are liked and accepted by more and more people. With the development of globalization, more and more domestic audiences like to watch foreign movies, especially young people. Domestic movies are gradually going abroad and are recognized and loved by the world. When watching TV movies in different languages, subtitle translation is extremely important. High-quality subtitle translation can add icing on the cake for TV movies. On the contrary, if the subtitle translation is unsatisfactory, it will reduce the quality of TV movies, affect the audience's viewing experience, and also have a serious impact on the subsequent application as a corpus resource. Therefore, how to efficiently detect the translation quality of video subtitles has become an urgent problem to be solved. Summary of the invention

[0003] The technical problem to be solved by the present invention is that there is currently no method for quality detection of subtitles translated from film and television works, and the quality of subtitle translation cannot be guaranteed. Therefore, the present invention provides a subtitle translation quality detection method, device, equipment and medium to perform quality detection on video subtitles, ensure high-quality video subtitles, and subsequently provide a reliable basis as a corpus resource.

[0004] The present invention is achieved through the following technical solutions:

[0005] A subtitle translation quality detection method, comprising:

[0006] Obtaining a bilingual subtitle translation video, performing speech recognition and image recognition on the bilingual subtitle translation video, and obtaining a speech recognition result and an image recognition result;

[0007] Filtering the original translation from the image recognition result, and processing the speech recognition result and the original translation into an original bilingual sentence pair according to the subtitle timeline;

[0008] Perform quality inspection on the original bilingual sentence pairs and obtain quality scores;

[0009] When the obtained quality score is lower than the score threshold, the original bilingual sentence pairs below the score threshold are corrected by the sentence pair correction model to obtain corrected bilingual sentence pairs;

[0010] The corrected bilingual sentence pairs are quality checked again and a new quality score is obtained. When the new quality score is still lower than the score threshold, the corrected bilingual sentence pairs are used as new original bilingual sentence pairs, and the steps of correcting the original bilingual sentence pairs that are lower than the score threshold by using the sentence pair correction model are repeatedly performed until the quality score is no lower than the score threshold.

[0011] Furthermore, the filtering out the original translation from the image recognition result includes:

[0012] The speech recognition results and the image recognition results are subjected to language identification, and the text content in the image recognition results that is inconsistent with the language of the speech recognition results is used as the original translation.

[0013] Furthermore, the quality detection of the original bilingual sentence pairs to obtain a quality score includes:

[0014] Performing machine translation on the speech recognition results in the original bilingual sentence pairs to obtain machine-translated texts;

[0015] The machine-translated text is used as the reference translation, and the BLEU algorithm is used to check the quality of the original translation to obtain a quality score.

[0016] Furthermore, the sentence pair correction model is used to correct the original bilingual sentence pairs below the scoring threshold, including:

[0017] The sentences in the original bilingual sentence pairs are merged and split by using the sentence pair correction model, and homophones are replaced in the merged and split original texts, and similar characters are replaced in the merged and split original translations, so as to obtain an initial corrected bilingual sentence pair;

[0018] Based on the dependency syntax, the completeness of the sentences in the bilingual sentence pairs after the initial correction is judged to complete the secondary correction process.

[0019] Furthermore, the merging and splitting of the sentences in the original bilingual sentence pairs includes:

[0020] The sentences in the original bilingual sentence pair are merged according to sentence pause markers, and the sentences in the original bilingual sentence pair are split according to the meaning group division rule.

[0021] Furthermore, in the step of repeatedly performing correction processing on original bilingual sentence pairs below the scoring threshold by using the sentence pair correction model, the subtitle translation quality detection method further includes:

[0022] When repeatedly executing the step of correcting the original bilingual sentence pairs below the scoring threshold by the sentence pair correction model, if the quality score obtained within a preset number of iterations is not lower than the scoring threshold, then stopping the correction when the quality score is not lower than the scoring threshold;

[0023] When repeatedly executing the step of correcting the original bilingual sentence pairs below the scoring threshold through the sentence pair correction model, if the preset number of iterations has been reached but the obtained quality score is still below the scoring threshold, the correction is stopped and the original bilingual sentence pairs are sent to the subtitle verifier for verification.

[0024] A subtitle translation quality detection device, comprising:

[0025] A translation video processing module is used to obtain a bilingual subtitle translation video, perform speech recognition and image recognition on the bilingual subtitle translation video, and obtain speech recognition results and image recognition results;

[0026] A bilingual sentence pair acquisition module, used to filter out the original text and the original translation from the image recognition result, and process the speech recognition result and the original translation into the original bilingual sentence pair according to the subtitle timeline;

[0027] A primary quality detection module is used to perform quality detection on the original bilingual sentence pairs and obtain quality scores;

[0028] A correction processing module is used for correcting the original bilingual sentence pairs below the score threshold by using a sentence pair correction model to obtain a corrected bilingual sentence pair when the obtained quality score is below the score threshold;

[0029] The secondary quality detection module is used to perform quality detection on the corrected bilingual sentence pairs again and obtain a new quality score. When the new quality score is still lower than the scoring threshold, the corrected bilingual sentence pairs are used as new original bilingual sentence pairs, and the steps of correcting the original bilingual sentence pairs that are lower than the scoring threshold by the sentence pair correction model are repeatedly executed until the quality score is no lower than the scoring threshold.

[0030] Furthermore, the correction processing module includes:

[0031] An initial correction unit, used to merge and split the sentences in the original bilingual sentence pair through the sentence pair correction model, and replace homophones in the merged and split original texts, and replace similar characters in the merged and split original translations, so as to obtain an initially corrected bilingual sentence pair;

[0032] The secondary correction unit is used to judge the completeness of the sentences in the bilingual sentence pair after the initial correction based on the dependency syntax to complete the secondary correction process.

[0033] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned subtitle translation quality detection method when executing the computer program.

[0034] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the subtitle translation quality detection method is implemented.

[0035] The invention provides a subtitle translation quality detection method, device, equipment and medium. The method comprises the following steps: obtaining a bilingual subtitle translation video, performing speech recognition and image recognition on the bilingual subtitle translation video, obtaining a speech recognition result and an image recognition result; selecting an original translation from the image recognition result, processing the speech recognition result and the original translation into an original bilingual sentence pair according to a subtitle timeline, then performing quality detection on the original bilingual sentence pair to obtain a quality score. When the obtained quality score is lower than a score threshold, a sentence pair correction model is used to correct the original bilingual sentence pair that is lower than the score threshold to obtain a corrected bilingual sentence pair. The method further performs quality detection on the corrected bilingual sentence pair to obtain a new quality score. When the obtained new quality score is still lower than the score threshold, the corrected bilingual sentence pair is used as a new original bilingual sentence pair, and the step of correcting the original bilingual sentence pair that is lower than the score threshold by the sentence pair correction model is repeatedly performed until the obtained quality score is not lower than the score threshold, so as to complete the detection of the subtitle translation quality, determine the quality of the subtitle translation, and ensure the reliability of the subsequent use as a corpus resource. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0037] Figure 1 The present invention is a flowchart of a subtitle translation quality detection method.

[0038] Figure 2 for Figure 1 A specific flow chart of step S40 in FIG.

[0039] Figure 3 The present invention is a schematic diagram of the structure of a subtitle translation quality detection device.

[0040] Figure 4 A schematic diagram of a computer device of the present invention. DETAILED DESCRIPTION

[0041] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0042] The present invention provides a subtitle translation quality detection method, which can be applied to different electronic devices, including but not limited to various personal computers, laptops, smart phones and tablet computers.

[0043] In one embodiment, if Figure 1 As shown, the present invention provides a subtitle translation quality detection method, comprising:

[0044] S10: Obtain a bilingual subtitle translation video, perform speech recognition and image recognition on the bilingual subtitle translation video, and obtain a speech recognition result and an image recognition result;

[0045] S20: Filtering the original translation from the image recognition result, and processing the speech recognition result and the original translation into an original bilingual sentence pair according to the subtitle timeline;

[0046] S30: Perform quality detection on the original bilingual sentence pairs to obtain quality scores;

[0047] S40: When the obtained quality score is lower than the score threshold, the original bilingual sentence pair below the score threshold is corrected by the sentence pair correction model to obtain a corrected bilingual sentence pair;

[0048] S50: Perform quality inspection on the corrected bilingual sentence pair again and obtain a new quality score. When the new quality score is still lower than the score threshold, the corrected bilingual sentence pair is used as the new original bilingual sentence pair, and the step of correcting the original bilingual sentence pair that is lower than the score threshold by using the sentence pair correction model is repeatedly performed until the quality score is no lower than the score threshold.

[0049] Among them, bilingual subtitle translation videos refer to videos that include original subtitles and translated subtitles.

[0050] The original bilingual sentence pairs refer to the sentence pairs in two languages ​​formed by the speech recognition results and the original translation according to the subtitle timeline.

[0051] The sentence pair correction model refers to a model for correcting original bilingual sentence pairs whose quality scores are lower than the score threshold. The sentence pair correction model includes merging, splitting, replacing homophones and similar characters, and judging sentence completeness of the original bilingual sentence pairs.

[0052] As an example, in step S10, a bilingual subtitle translation video is obtained, and speech recognition technology is used to recognize the speech of the speaker in the obtained bilingual subtitle translation video, and the recognized text content is obtained as the speech recognition result. Image recognition technology is used to perform image recognition on the subtitles in each frame of the bilingual subtitle translation video, and the recognized text content is obtained as the image recognition result. The speech recognition technology and image recognition technology in this embodiment adopt existing technologies and are not introduced here.

[0053] As an example, in step S20, after obtaining the speech recognition result and the image recognition result, since the speech recognition result is obtained by recognizing the speaker's speech, the language of the speech recognition result is consistent with the original text. Since the present embodiment uses a bilingual subtitle translation video, after obtaining the image recognition result, the image recognition result includes both the original text and the translation, therefore, it is necessary to filter out the text content that is inconsistent with the language of the image recognition result from the language of the speech recognition result as the original translation.

[0054] Since the subtitles in the bilingual subtitle translation video are synchronized with the speaker's voice, after obtaining the original translation from the image recognition result, the speech recognition result and the original translation are matched one by one according to the timestamp on the subtitle timeline to form a bilingual sentence pair.

[0055] As an example, in step S30, after obtaining the speech recognition result, the speech recognition result is machine translated to obtain a machine-translated text, and the machine-translated text is used as a reference translation, and the BLUE algorithm is used to perform quality detection on the original translation to obtain a quality score.

[0056] As an example, in step S40, after obtaining the quality score of the original translation, the obtained quality score is compared with a preset scoring threshold. When the obtained quality score is lower than the scoring threshold, it means that the speech recognition result is not accurate enough or the translation quality of the original translation cannot meet the requirements. The sentence pair correction model is required to correct the speech recognition results and the original translation in the original bilingual sentence pairs that are lower than the scoring threshold to obtain a corrected sentence pair.

[0057] As an example, in step S50, after obtaining the corrected bilingual sentence pair, the corrected bilingual sentence pair is again quality-checked and a new quality score is obtained, and the new quality score is compared with the score threshold. When the obtained quality score is lower than the score threshold, it means that the corrected bilingual sentence pair is still inaccurate. Therefore, the corrected bilingual sentence pair is used as the new original bilingual sentence pair, and the step of correcting the sentence pair below the score threshold by the sentence pair correction model is repeatedly performed until the obtained quality score is not lower than the score threshold. When repeatedly performing the step of correcting the original bilingual sentence pair below the score threshold by the sentence pair correction model, if the quality score obtained within the preset number of iterations is not lower than the score threshold, the correction is stopped when the quality score is not lower than the score threshold; if the preset number of iterations has been reached but the obtained quality score is still lower than the score threshold, the correction is stopped and the original bilingual sentence pair is sent to the subtitle checker for verification.

[0058] In one embodiment, if Figure 2As shown in the figure, in step S40, the sentence pair correction model is used to correct the translation pairs below the scoring threshold, which specifically includes the following steps:

[0059] S41: The sentence pair correction model merges and splits the sentences in the original bilingual sentence pair, replaces homophonic characters in the merged and split original text, and replaces similar-shaped characters in the merged and split original translation, to obtain the initially corrected bilingual sentence pair;

[0060] S42: Based on dependency syntax, the integrity of the sentences in the initially corrected bilingual sentence pair is judged to complete the secondary correction process.

[0061] Among them, dependency syntax was first proposed by the French linguist L. Tesniere, which is a method of analyzing a sentence into a dependency syntax tree to describe the dependency relationships between words.

[0062] As an example, in step S41, the sentence pair correction model is used to merge the sentences in the original bilingual sentence pair according to the sentence pause markers, and split the sentences in the original bilingual sentence pair according to the sense group division rules to complete the merging and splitting of the sentences. The sentences after the merging and splitting are detected to obtain the characters at the interruption of the sentence as the interrupted characters and the words at the interruption as the interrupted words. The dictionary is queried based on the interrupted characters or interrupted words to determine whether there are homophonic characters or similar-shaped characters. If there are homophonic characters or similar-shaped characters, the interrupted characters or interrupted words are replaced with homophonic characters or similar-shaped characters to obtain the initially corrected sentences.

[0063] Furthermore, since the machine translation is obtained by translating the speech recognition results, only homophonic characters may appear in the speech recognition results, that is, one pronunciation corresponds to multiple characters (such as "秘密"); the original translation is obtained by image recognition, and only similar-shaped characters may appear during image recognition, that is, two characters are similar in shape (such as "诲" and "海", "烂" and "栏"). Therefore, if it is the original text in the sentence pair, homophonic characters are replaced; if it is the original translation in the sentence pair, similar-shaped characters are replaced.

[0064] Specifically, the process of replacing the interrupted characters or interrupted words with homophonic characters or similar-shaped characters specifically includes:

[0065] The first step: Replace the homophonic characters in the speech recognition results of the original bilingual sentence pair, or replace the similar-shaped characters in the original translation of the original bilingual sentence pair, to obtain the initially corrected bilingual sentence pair. If the quality score of the initially corrected bilingual sentence pair is still lower than the scoring threshold, then execute the second step;

[0066] Step 2: If the first step performs homophone replacement, then this step performs similar-looking character replacement. If the first step performs similar-looking character replacement, then this step performs homophone replacement to obtain an initially corrected bilingual sentence pair. If the quality score of the initially corrected bilingual sentence pair is still lower than the score threshold, then the third step is performed.

[0067] Step 3: Simultaneously, the speech recognition results in the original bilingual sentence pairs are replaced with homophones, and the original translations in the original bilingual sentence pairs are replaced with similar characters to obtain the initially corrected bilingual sentence pairs.

[0068] As an example, the specific process of splitting sentences according to the idea group segmentation rules in this embodiment is: (1) calling the idea group segmentation script to divide the translated text into sentences according to the sentence segmentation marks; (2) using the part-of-speech analysis tool to perform part-of-speech tagging on each word in the translated sentence; (3) dividing the translated sentence into ideas according to the preset idea group segmentation rules.

[0069] The segmentation mark refers to a mark that can be used to segment text content in any language, and is not limited here. Taking Chinese as an example, the segmentation mark can be a punctuation mark ".", "!", "?", ":", or other forms of identifiers.

[0070] The idea group segmentation script refers to a script that divides sentences according to the text to be translated, and divides the divided sentences into idea groups according to the idea group segmentation rules. The idea group segmentation rules refer to the rules for segmenting sentences according to the sentence grammar. The idea group segmentation rules in this embodiment include but are not limited to subject, predicate and object, prepositional phrases as object complement, verb phrase + object, verb phrase + gerund phrase object, prepositional phrase as adverbial, noun phrase as subject, predicate + object, participle phraseization, adverbial conjunction, preposition + participle phrase as adverbial, and prepositional phrase postpositioned attributive.

[0071] Among them, a phrase refers to a word group that is relatively complete in meaning, closely related in grammar, and cannot be divided any further. A phrase group can be composed of a word, a word group or a phrase, or a clause, or a certain grammatical structure. In a sentence, a phrase group is internally manifested as a close grammatical relationship, and externally manifested as a phrase group that cannot be paused in the middle of the pronunciation. A phrase group cannot be split arbitrarily, otherwise it will cause misunderstanding. This embodiment utilizes the unique properties of the phrase group itself to perform phrase segmentation on the translation text to obtain a phrase text.

[0072] As an example, in step S42, after obtaining the initially corrected sentence, the completeness of the sentence in the initially corrected bilingual sentence pair is judged based on the dependency syntax to complete the secondary correction process.

[0073] As an example, in step S50, the specific process of performing quality inspection on the corrected bilingual sentence pair again and obtaining a new quality score is as follows:

[0074] If the correction is to replace homophones in the speech recognition results of the original bilingual sentence pairs, the corrected speech recognition results are machine translated to obtain a new machine-translated text, and the new machine-translated text is used as a reference translation. The BLEU algorithm is used to perform quality inspection on the original translation to obtain a new quality score;

[0075] If the original translation in the original bilingual sentence pair is replaced with similar characters during the revision, the machine-translated text of the speech recognition result in the original bilingual sentence pair is used as the reference translation, and the BLEU algorithm is used to perform quality inspection on the revised translation to obtain a new quality score;

[0076] If during the revision, both the original translation in the original bilingual sentence pair is replaced with similar characters, and the speech recognition result in the original bilingual sentence pair is replaced with homophones, then the revised speech recognition result is machine translated to obtain a new machine-translated text, and the new machine-translated text is used as a reference translation. The BLEU algorithm is used to perform quality inspection on the revised translation to obtain a new quality score.

[0077] The invention provides a subtitle translation quality detection method. The method performs speech recognition and image recognition on a bilingual subtitle translation video to obtain speech recognition results and image recognition results. Then, based on the language of the speech recognition result, an original translation is screened out from the image recognition result. According to a subtitle timeline, the speech recognition result and the original translation are processed into a bilingual sentence pair. Then, the speech recognition result is machine translated to obtain a machine-translated text. The machine-translated text is used as a reference translation. A BLEU algorithm is used to perform quality detection on the original translation to obtain a quality score. When the obtained quality score is lower than a score threshold, a sentence pair correction model is used to correct the original bilingual sentence pair that is lower than the score threshold to obtain a corrected bilingual sentence pair. The corrected bilingual sentence pair is again quality detected to obtain a new quality score. When the obtained new quality score is still lower than the score threshold, the step of correcting the original bilingual sentence pair that is lower than the score threshold by the sentence pair correction model is repeatedly performed until the obtained quality score is not lower than the score threshold, so as to complete the detection of the subtitle translation quality, ensure high-quality video translation subtitles, and provide a reliable data source for subsequent use as corpus resources.

[0078] In one embodiment, a subtitle translation quality detection device is provided, and the subtitle translation quality detection device corresponds one-to-one to a subtitle translation quality detection method in the above embodiment. Figure 3 As shown, the subtitle translation quality detection device includes a translation video processing module 10, a bilingual sentence pair acquisition module 20, a primary quality detection module 30, a correction processing module 40 and a secondary quality detection module 50. The functional modules are described in detail as follows:

[0079] The translation video processing module 10 is used to obtain a bilingual subtitle translation video, perform speech recognition and image recognition on the bilingual subtitle translation video, and obtain speech recognition results and image recognition results;

[0080] A bilingual sentence pair acquisition module 20 is used to filter out the original text and the original translation from the image recognition result, and process the speech recognition result and the original translation into the original bilingual sentence pair according to the subtitle timeline;

[0081] A primary quality detection module 30, used to perform quality detection on the original bilingual sentence pairs and obtain a quality score;

[0082] A correction processing module 40 is used to correct the original bilingual sentence pairs below the score threshold by using a sentence pair correction model to obtain a corrected bilingual sentence pair when the obtained quality score is below the score threshold;

[0083] The secondary quality detection module 50 is used to perform quality detection on the corrected bilingual sentence pair again and obtain a new quality score. When the new quality score is still lower than the score threshold, the corrected bilingual sentence pair is used as the new original bilingual sentence pair, and the step of correcting the original bilingual sentence pair below the score threshold by the sentence pair correction model is repeatedly executed until the quality score is no lower than the score threshold.

[0084] Furthermore, the correction processing module 40 includes an initial correction unit and a secondary correction unit.

[0085] An initial correction unit, used to merge and split the sentences in the original bilingual sentence pair through the sentence pair correction model, and replace homophones in the merged and split original texts, and replace similar characters in the merged and split original translations, so as to obtain an initially corrected bilingual sentence pair;

[0086] The secondary correction unit is used to judge the completeness of the sentences in the bilingual sentence pair after the initial correction based on the dependency syntax to complete the secondary correction process.

[0087] The specific definition of a subtitle translation quality detection device can be found in the definition of a subtitle translation quality detection method mentioned above, which will not be repeated here. Each module in the above-mentioned subtitle translation quality detection device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0088] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 4As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a computer-readable storage medium and an internal memory. The computer-readable storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the computer-readable storage medium. The database of the computer device is used to store data involved in a subtitle translation quality detection method. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a subtitle translation quality detection method is implemented.

[0089] A computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a subtitle translation quality detection method in the above embodiment are implemented, such as Figure 1 Steps S10-S50 shown, or Figure 2 To avoid repetition, the steps shown in are not described here. Alternatively, when the processor executes the computer program, the functions of each module / unit of the subtitle translation quality detection device in the above embodiment are realized, such as Figure 3 The functions of the modules 10 to 50 are shown in Figure 1. To avoid repetition, they will not be described here.

[0090] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a subtitle translation quality detection method in the above embodiment are implemented, such as Figure 1 Steps S10-S50 shown, or Figure 2 To avoid repetition, the steps shown in are not described here. Alternatively, when the processor executes the computer program, the functions of each module / unit in this embodiment of a subtitle translation quality detection device are realized, such as Figure 3 The functions of the modules 10 to 50 are shown in Figure 1. To avoid repetition, they will not be described here.

[0091] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0092] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0093] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A subtitle translation quality detection method, characterized in that: include: Obtaining a bilingual subtitle translation video, performing speech recognition and image recognition on the bilingual subtitle translation video, and obtaining a speech recognition result and an image recognition result; Filtering the original translation from the image recognition result, and processing the speech recognition result and the original translation into an original bilingual sentence pair according to the subtitle timeline; Perform quality inspection on the original bilingual sentence pairs and obtain quality scores; When the obtained quality score is lower than the score threshold, the original bilingual sentence pairs below the score threshold are corrected by the sentence pair correction model to obtain corrected bilingual sentence pairs; The corrected bi-sentence pairs are quality-checked again and a new quality score is obtained. If the new quality score is still lower than the scoring threshold, the corrected bi-sentence pairs are used as new original bi-sentence pairs, and the steps of correcting the original bi-sentence pairs that are lower than the scoring threshold by using the sentence pair correction model are repeated until the quality score is no lower than the scoring threshold. The sentence pair correction model is used to correct the original bilingual sentence pairs below the scoring threshold, including: The sentences in the original bilingual sentence pairs are merged and split by using the sentence pair correction model, and homophones are replaced in the merged and split original texts, and similar characters are replaced in the merged and split original translations, so as to obtain an initial corrected bilingual sentence pair; Based on dependency syntax, the completeness of the sentences in the initially corrected bilingual sentence pairs is judged to complete the secondary correction process; The merging and splitting of the sentences in the original bilingual sentence pairs includes: The sentences in the original bilingual sentence pairs are merged according to sentence pause markers, and the sentences in the original bilingual sentence pairs are split according to the meaning group division rule.

2. A subtitle translation quality detection method according to claim 1, characterized in that: The image recognition Filter out the original translation from other results, including: The speech recognition results and the image recognition results are subjected to language identification, and the text content in the image recognition results that is inconsistent with the language of the speech recognition results is used as the original translation.

3. A subtitle translation quality detection method according to claim 1, characterized in that: The quality detection of the original bilingual sentence pairs to obtain the quality score includes: Performing machine translation on the speech recognition results in the original bilingual sentence pairs to obtain machine-translated texts; The machine-translated text is used as the reference translation, and the BLEU algorithm is used to check the quality of the original translation to obtain a quality score.

4. A subtitle translation quality detection method according to claim 1, characterized in that: In the repeated execution of the sentence In the step of using the correction model to correct the original bilingual sentence pairs below the scoring threshold, the subtitle translation quality check The testing methods also include: When repeatedly executing the step of correcting the original bilingual sentence pairs below the scoring threshold by the sentence pair correction model, if the quality score obtained within a preset number of iterations is not lower than the scoring threshold, then stopping the correction when the quality score is not lower than the scoring threshold; When repeatedly executing the step of correcting the original bilingual sentence pairs below the scoring threshold through the sentence pair correction model, if the preset number of iterations has been reached but the obtained quality score is still below the scoring threshold, the correction is stopped and the original bilingual sentence pairs are sent to the subtitle verifier for verification.

5. A subtitle translation quality detection device, characterized in that: include: A translation video processing module is used to obtain a bilingual subtitle translation video, perform speech recognition and image recognition on the bilingual subtitle translation video, and obtain speech recognition results and image recognition results; A bilingual sentence pair acquisition module, used to filter out the original text and the original translation from the image recognition result, and process the speech recognition result and the original translation into the original bilingual sentence pair according to the subtitle timeline; A primary quality detection module is used to perform quality detection on the original bilingual sentence pairs and obtain quality scores; A correction processing module is used for correcting the original bilingual sentence pairs below the score threshold by using a sentence pair correction model to obtain a corrected bilingual sentence pair when the obtained quality score is below the score threshold; A secondary quality detection module is used to perform quality detection on the corrected bi-sentence pairs again and obtain a new quality score. When the new quality score is still lower than the score threshold, the corrected bi-sentence pairs are used as new original bi-sentence pairs, and the steps of correcting the original bi-sentence pairs that are lower than the score threshold by using the sentence pair correction model are repeatedly performed until the quality score is no lower than the score threshold. The correction processing module comprises: An initial correction unit, used to merge and split the sentences in the original bilingual sentence pair through the sentence pair correction model, and replace homophones in the merged and split original texts, and replace similar characters in the merged and split original translations, so as to obtain an initially corrected bilingual sentence pair; A secondary correction unit, used for judging the completeness of the sentences in the bilingual sentence pair after the initial correction based on dependency syntax, and completing the secondary correction process; The merging and splitting of the sentences in the original bilingual sentence pairs includes: The sentences in the original bilingual sentence pairs are merged according to sentence pause markers, and the sentences in the original bilingual sentence pairs are split according to the meaning group division rule.

6. A computer device comprising a memory, a processor, and a computer program product stored in the memory and operable in the processor A computer program running on a computer, characterized in that When the processor executes the computer program, the subtitle translation quality detection method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the subtitle translation quality detection method according to any one of claims 1 to 4 is implemented.

Images