Method and apparatus for predicting pronunciation of polyphonic character

Abstract

This specification provides embodiments of a method and apparatus for predicting the pronunciation of polyphonic characters. The method includes: obtaining interactive content to be played to a user; inputting the interactive content into a pre-trained prosodic prediction model, which outputs interactive content with prosodic tags; wherein the prosodic tags indicate pauses during pronunciation; for each content unit segmented by the prosodic tags in the interactive content, detecting whether the content unit contains a polyphonic character; if so, determining the pronunciation of the polyphonic character in the content unit based on its context. This specification enables machines to accurately determine the pronunciation of polyphonic characters and improves the processing efficiency of machines.

Description

TECHNICAL FIELD

[0001] One or more embodiments of the present specification relate to network communication technology, and in particular to a multi-sound character pronunciation prediction method and device. BACKGROUND

[0002] At present, there are more and more human-computer interaction businesses. Machine devices can broadcast corresponding dialogues according to the dialogue content of the user. In a real intelligent dialogue, it is impossible to synthesize voice in advance in the machine device, and the machine device needs to generate and broadcast the voice in real time according to the dialogue content sent by the user in real time.

[0003] There are multi-sound characters in Chinese characters. One Chinese character can have multiple pinyins, that is, multiple pronunciations, such as "Xing" and "de". The pronunciation of multi-sound characters has no uniform regularity. This leads to that in human-computer interaction, when the machine device needs to play a segment of voice, if there is a multi-sound character in the segment of voice, the machine device is difficult to determine the accurate pronunciation of the multi-sound character in the current segment of voice to be played.

[0004] Therefore, a more effective multi-sound character pronunciation prediction method suitable for machine devices is needed. SUMMARY

[0005] One or more embodiments of the present specification describe a multi-sound character pronunciation prediction method and device, which can enable the machine device to accurately determine the pronunciation of the multi-sound character, so that the played voice is more natural and more convenient for the user to understand.

[0006] According to a first aspect, a multi-sound character pronunciation prediction method is provided, wherein the method comprises:

[0007] Obtaining interactive content that needs to be played to a user;

[0008] Inputting the interactive content into a pre-trained prosody prediction model, and outputting the interactive content with prosody labels from the prosody prediction model; wherein the prosody label is used to indicate a pause when pronouncing;

[0009] For each content unit segmented by each prosody label in the interactive content, detecting whether the content unit includes a multi-sound character, and if so, obtaining the pronunciation of the included multi-sound character in the content unit according to the context of the content unit.

[0010] The pronunciation of the included multi-sound character in the content unit according to the context of the content unit comprises: inputting the content unit into a pre-trained multi-sound character prediction model to obtain the pronunciation of each character in the content unit including the multi-sound character output by the multi-sound character prediction model according to the context of the content unit.

[0011] The method further comprises: presetting a monosyllabic word dictionary; and the monosyllabic word dictionary comprises pronunciation of each monosyllabic word.

[0012] After detecting that the content unit does not comprise a polyphonic word, the method further comprises:

[0013] According to the preset monosyllabic word dictionary, pronunciation of each word in the content unit is obtained.

[0014] The method further comprises: playing the interactive content to the user in a prosody-carrying voice mode according to prosody labels carried in the interactive content and the obtained pronunciation of each word including the polyphonic word.

[0015] The method further comprises: after detecting that a content unit comprises only one word and the word is a polyphonic word, before obtaining pronunciation of the included polyphonic word in the content unit, the method further comprises: merging the content unit in which the polyphonic word is located and an adjacent content unit into one content unit.

[0016] The method further comprises: inputting the merged content unit into a pre-trained polyphonic word prediction model to obtain pronunciation of each word including the polyphonic word in the merged content unit output by the polyphonic word prediction model according to a context of the merged content unit.

[0017] The training method of the prosody prediction model comprises:

[0018] Obtaining text corpus in a target business scenario;

[0019] Performing normalization processing on the text corpus;

[0020] Obtaining sample corpus; wherein the sample corpus is obtained by marking the normalized text corpus with prosody labels; and the prosody labels are used to indicate pause duration.

[0021] Initializing model structure and parameters of the prosody prediction model by using model structure and parameters of the punctuation prediction model after training.

[0022] Inputting the sample corpus with the prosody labels into the initialized prosody prediction model to train the prosody prediction model.

[0023] The prosody labels comprise at least one of the following: a label indicating word segmentation, a label indicating a small pause, a label indicating a large pause, and a label indicating end of a sentence.

[0024] According to a second aspect, a polyphonic word pronunciation prediction device is provided, wherein the device comprises:

[0025] The interactive content acquisition module is configured to obtain the interactive content that needs to be played to the user;

[0026] The prosody processing module is configured to input the interactive content into a pre-trained prosody prediction model, and the prosody prediction model outputs interactive content with prosody tags; wherein, the prosody tags are used to indicate pauses during pronunciation.

[0027] The pronunciation processing module is configured to detect whether each content unit segmented by each prosody tag in the interactive content contains a polyphonic character. If it does, the module obtains the pronunciation of the polyphonic character in the content unit based on the context of the content unit.

[0028] According to a third aspect, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.

[0029] According to a fourth aspect, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor, when executing the executable code, implements the method described in any embodiment of this specification.

[0030] The method and apparatus for predicting the pronunciation of polyphonic characters provided in this specification first use a prosodic prediction model to add prosodic tags to a complete segment of content to be played. These prosodic tags indicate which words in the segment should be followed by a pause; in other words, by adding prosodic tags, a complete segment of content, i.e., a long sentence, is effectively divided into multiple content units. For each segmented content unit, the characters within that unit are semantically strongly related, allowing each character to more accurately represent the correct meaning within that unit. Therefore, this specification uses each content unit segmented by prosodic tags as a pronunciation prediction unit to predict the pronunciation of polyphonic characters within that unit. This avoids requiring too much contextual information when predicting the pronunciation of polyphonic characters; instead, it only needs to refer to the context within semantically related content units. Therefore, the prediction of polyphonic character pronunciation is more accurate, and the processing speed is greatly improved. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the system architecture used in one embodiment of this specification.

[0033] Figure 2 This is a flowchart of a method for predicting the pronunciation of polyphonic characters in one embodiment of this specification.

[0034] Figure 3 This is a flowchart of a training method for a prosody prediction model in one embodiment of this specification.

[0035] Figure 4 This is a schematic diagram of the pronunciation prediction process for polyphonic characters in one embodiment of this specification.

[0036] Figure 5 This is a flowchart of a method for predicting the pronunciation of polyphonic characters in another embodiment of this specification.

[0037] Figure 6 This is a schematic diagram of the structure of a device for predicting the pronunciation of polyphonic characters in one embodiment of this specification. Detailed Implementation

[0038] The solution provided in this specification will now be described with reference to the accompanying drawings.

[0039] First, it should be noted that the terminology used in the embodiments of this invention is for the purpose of describing specific embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0040] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0041] To facilitate understanding of the methods provided in this manual, the system architecture involved and applicable to this manual is first described. For example... Figure 1 As shown, the system architecture mainly includes users and machine devices for human-computer interaction.

[0042] The machines and devices that interact with users can include, but are not limited to, smart mobile terminals, smart home devices, network devices, wearable devices, smart medical devices, and PCs (personal computers). Smart mobile devices can include mobile phones, tablets, laptops, PDAs (personal digital assistants), and connected cars. Smart home devices can include smart appliances such as smart TVs, smart air conditioners, smart water heaters, smart refrigerators, and smart air purifiers; they can also include smart door locks, smart sockets, smart lights, and smart cameras. Network devices can include switches, wireless access points (APs), and servers. Wearable devices can include smartwatches, smart glasses, smart bracelets, virtual reality devices, augmented reality devices, and mixed reality devices (devices that support both virtual and augmented reality). Smart medical devices can include smart thermometers, smart blood pressure monitors, and smart blood glucose meters.

[0043] It should be understood that Figure 1 The number of users and machines shown is merely illustrative. Any number can be selected and deployed as needed.

[0044] Figure 2 This is a flowchart of a method for predicting the pronunciation of polyphonic characters in one embodiment of this specification. The method is executed by a device for predicting the pronunciation of polyphonic characters, which can be installed in a machine or device requiring human-computer interaction. It is understood that this method can also be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. See also... Figure 2 The method includes:

[0045] Step 201: Obtain the interactive content that needs to be played to the user.

[0046] Step 203: Input the interactive content into the pre-trained prosody prediction model, and the prosody prediction model outputs interactive content with prosodic labels; wherein, the prosodic labels are used to indicate pauses during pronunciation.

[0047] Step 205: For each content unit segmented by each prosody tag in the interactive content, detect whether the content unit contains a polyphonic character. If it does, obtain the pronunciation of the polyphonic character in the content unit based on the context of the content unit.

[0048] It can be seen that, in Figure 2In the polyphonic character pronunciation prediction method of the embodiments shown in this specification, the entire segment of content to be played by the machine is not used as a prediction unit for polyphonic character pronunciation prediction. This is because such a method would lead to inaccurate prediction and low efficiency. For example, if a segment of content consists of 5 lines of text, meaning the text is very long and has no punctuation, then using these 5 lines of text as a prediction unit to predict the pronunciation of a polyphonic character in those 5 lines would require too much contextual information, resulting in inaccurate prediction and slow processing speed.

[0049] exist Figure 2 In the polyphonic character pronunciation prediction method of the embodiments shown in this specification, a prosodic prediction model is first used to add prosodic tags to a complete segment of content to be played. These prosodic tags indicate which words in the segment should be followed by a pause; in other words, adding prosodic tags effectively divides the complete segment, i.e., a long sentence, into multiple content units. For each segmented content unit, the characters within that unit are semantically strongly related, allowing each character to more accurately represent the correct meaning within that unit. Therefore, this embodiment uses each content unit segmented by prosodic tags as a pronunciation prediction unit to predict the pronunciation of polyphonic characters within that unit. This avoids requiring too much contextual information when predicting polyphonic character pronunciation; only the context within semantically related content units needs to be considered. Therefore, the prediction of polyphonic character pronunciation is more accurate, and the processing speed is greatly improved.

[0050] As mentioned earlier, the embodiments in this specification require a prosody prediction model to add prosody tags to interactive content. The following will first describe the implementation method of the prosody prediction model, and then describe how to use this prosody prediction model to achieve… Figure 2 The method for predicting the pronunciation of polyphonic characters shown is illustrated.

[0051] Human speech has rhythm, including the pauses after words and their duration. In human-computer interaction, to make the speech produced by machines more natural and easier to understand, machines also need to perform prosodic processing on the real-time generated speech content. This is especially true when the content to be played is long and lacks punctuation marks to indicate pauses. In such cases, machines need to use prosodic prediction models to label the interactive content with prosodic tags. These tags then guide the machine to implement rhythmic pauses, resulting in more natural-sounding speech.

[0052] In the embodiments described in this specification, the prosody prediction model may include the following two implementation methods:

[0053] Method 1: Use word segmentation capabilities to add prosodic tags.

[0054] Method 2: Add prosodic tags by utilizing the predictive capabilities of punctuation marks.

[0055] In the first method described above, a prosodic prediction model with word segmentation capabilities can be pre-trained. The prosodic prediction model segments the content into words using various prosodic tags, and then performs prosodic pauses based on the words.

[0056] In Method Two above, a prosodic prediction model with punctuation prediction capabilities can be pre-trained. Considering that the punctuation prediction model can add punctuation marks to a text, and that punctuation marks are strongly correlated with prosodic rhythm (meaning that pauses are needed where punctuation marks are present), training the prosodic prediction model with the punctuation prediction model allows it to mark prosodic labels at the locations where various punctuation marks are needed in a text, indicating the need for pauses, thus enabling prosodic prediction.

[0057] Compared to Method 1, Method 2 does not add prosodic tags according to words. The prosodic prediction model determines which words, phrases, and sentences should be followed by punctuation marks in the past based on the training process. In fact, it determines which words, phrases, and sentences should be followed by prosodic pauses in the past. In this way, even if a sentence to be broadcast does not have punctuation marks and the text is long, the prosodic prediction model can still mark the long text with more accurate prosodic tags in subsequent prediction operations based on its training capabilities, making the broadcast speech more natural and not affecting the listener's understanding.

[0058] The following explains the method for training the prosody prediction model when using method two.

[0059] Figure 3 This is a flowchart illustrating a method for training a prosody prediction model in one embodiment of this specification. See also... Figure 3 The method includes:

[0060] Step 301: Obtain the text corpus in the target business scenario.

[0061] Step 303: Normalize the text corpus.

[0062] Step 305: Obtain sample corpus; wherein, the sample corpus is obtained by adding prosodic tags to the normalized text corpus; the prosodic tags are used to indicate the duration of pauses.

[0063] Step 307: Initialize the model structure and parameters of the prosody prediction model using the model structure and parameters of the trained punctuation prediction model.

[0064] Step 309: Input the sample corpus with prosody labels into the initialized prosody prediction model to train the prosody prediction model.

[0065] If a large amount of text corpus is manually labeled with prosody to obtain a large number of training samples, it can be understood that the process is extremely inefficient and costly. Referring to Figure 3 the method of the embodiment of the present specification shown, because the model structure and parameters of the punctuation prediction model are used, when training the prosody prediction model, there is no need for a large number of training samples that are manually labeled with prosody. That is to say, a prosody prediction model that meets business requirements can be trained using a small sample.

[0066] Next, each step of the Figure 3 shown process will be described separately.

[0067] First, for step 301: Obtain the text corpus in the target business scenario.

[0068] For example, if the target business scenario is a customer service scenario, such as handling customer complaints, then in this step 301, the text corpus related to customer service can be obtained for subsequent generation of training samples.

[0069] Next, for step 303: Perform normalization processing on the text corpus.

[0070] In this step 303, the normalization processing is for subsequent machine devices to recognize the text corpus. The normalization processing may include at least one of the following:

[0071] Remove non-Chinese character symbols in the text corpus that do not affect semantic understanding;

[0072] Convert non-Chinese character symbols in the text corpus that affect semantic understanding into Chinese characters with corresponding semantics, such as converting "￥" into the Chinese character "Renminbi".

[0073] Next, for step 305: Obtain the sample corpus; wherein, the sample corpus is obtained by attaching prosody labels to the normalized text corpus; the prosody labels are used to indicate the pause duration.

[0074] In the embodiments of this specification, prosodic tags are used to indicate the duration of pauses. For example, prosodic tag "#1" indicates word segmentation (word segmentation can correspond to the smallest pause duration), prosodic tag "#2" indicates a short pause (a short pause can correspond to a relatively short pause duration, such as 0.3 seconds), prosodic tag "#3" indicates a long pause (a long pause can correspond to a relatively long pause duration, such as 0.5 seconds), and prosodic tag "#4" indicates the end of a sentence (the end of a sentence can correspond to the longest pause duration, such as 1 second).

[0075] In the embodiments of this specification, the prosody tag can also be used to indicate the tone of the speech. For example, the prosody tag "#5" indicates playback with a rising tone, and the prosody tag "#6" indicates playback with a falling tone.

[0076] In step 305, manual labeling is performed, meaning that the normalized text corpus is manually labeled with prosodic tags. For example, if a piece of text is very long and has no punctuation marks, the manual labeler will mark which word in the long text should be followed by a prosodic tag. For example, which word in the long text should be followed by a prosodic tag "#1", which word should be followed by a prosodic tag "#3", which word should be followed by a prosodic tag "#5", and so on.

[0077] As mentioned above, since the prosody prediction model is trained on the basis of the punctuation prediction model in the embodiments of this specification, there is no need for a large number of training samples. In other words, only a small number of text corpora need to be manually labeled to generate a small number of training samples, such as a few hundred.

[0078] Next, for step 307: Initialize the prosody prediction model's structure and parameters using the trained punctuation prediction model's model structure and parameters.

[0079] Similar to the punctuation prediction model, the prosody prediction model includes a text preprocessing module, a Word3Vec module, and a BiLSTM module.

[0080] In the prosody prediction model, the text preprocessing module functions to perform normalization processing on the text corpus to obtain sample corpus.

[0081] The Word3Vec module includes functions for obtaining the embedding value of each character in the sample corpus;

[0082] The BiLSTM module's functions include outputting prediction results based on the embedding value of each character in the sample corpus. The specific categories of prosodic labels output by the BiLSTM module can be determined manually.

[0083] In one embodiment of this specification, during step 307, the prosodic prediction model can be fine-tuned during the initialization phase based on the differences between the functions of the prosodic prediction model and the punctuation prediction model. For example, if the punctuation prediction model outputs 4 categories of punctuation tags, while the prosodic prediction model outputs 5 categories of prosodic tags, then the initialization phase further includes: removing fully connected layers from the network structure of the punctuation prediction model; and setting the various categories output by the model to the categories included in the prosodic tags.

[0084] Next, for step 309: input the sample corpus with prosodic labels into the initialized prosodic prediction model to train the prosodic prediction model.

[0085] The initialized prosodic prediction model possesses punctuation prediction capabilities. When sample corpora with manually tagged prosodic labels are input into the initialized model, it can learn the correspondence between punctuation marks and prosodic labels. For example, because the initialized model has punctuation prediction capabilities, for the input sample corpus, if it determines that a comma should follow the Chinese character '1', and the manually tagged prosodic label is "#3", then the prosodic prediction model can output the prosodic label "#3" at every position where a comma should be added. Simultaneously, the prosodic prediction model can also learn how to perform prosodic labeling on text based on the manually tagged prosodic labels.

[0086] As can be seen, in the embodiments of this specification, the prosody prediction model simultaneously possesses the ability to: first predict punctuation marks, and then predict prosody tags based on the punctuation marks; and the ability to directly predict prosody tags based on the text content.

[0087] After multiple rounds of training, the prosody prediction model is complete.

[0088] The method for predicting the pronunciation of polyphonic characters will be described next. In one embodiment of this specification, see [link to embodiment]. Figure 4 and Figure 5 Methods for predicting the pronunciation of polyphonic characters include:

[0089] Step 501: Obtain the interactive content that needs to be played to the user.

[0090] In human-computer interaction, machines generate interactive content that needs to be played to users in real time based on their real-time interactions with them, such as generating answers to users' questions in real time.

[0091] For example, the interactive content includes "This disease is most common in winter."

[0092] Step 503: Input the interactive content into the pre-trained prosody prediction model, and the prosody prediction model outputs interactive content with prosodic labels; wherein, the prosodic labels are used to indicate pauses during pronunciation.

[0093] As mentioned earlier, if the prosodic prediction model is implemented using Method 1, it adds a prosodic tag to each word in the interactive content. If the prosodic prediction model is implemented using Method 2, it adds a prosodic tag to each prosodic content unit in the interactive content. A content unit segmented by each prosodic tag can be a single character, a word, or a phrase. For example, interactive content with prosodic tags might include "The disease #1 is most prevalent in #1 winter #4".

[0094] Step 505: For each content unit segmented by each prosody tag in the interactive content, detect whether the content unit contains polyphonic characters. If it does, proceed to step 509; otherwise, proceed to step 507.

[0095] A dictionary of polyphonic characters can be set in advance. In step 505, you can search the dictionary to see if each content unit contains polyphonic characters.

[0096] Step 507: Based on the pre-set monosyllabic dictionary, obtain the pronunciation of each character in the content unit, return to step 505 until all content units have been processed, and then execute step 517.

[0097] The monosyllabic dictionary includes the pronunciation of each monosyllabic character.

[0098] Step 509: Determine whether the content unit contains only one character and that character is a polyphonic character. If yes, proceed to step 511; otherwise, proceed to step 515.

[0099] Step 511: Merge the content unit containing the polyphonic character with the adjacent content units into one content unit.

[0100] Step 513: Input the merged content unit into the pre-trained polyphonic character prediction model to obtain the pronunciation of each character in the merged content unit, including polyphonic characters, output by the polyphonic character prediction model according to the context of the merged content unit. Return to step 505 until all content units have been processed and then execute step 517.

[0101] Step 515: Input the content unit into the pre-trained polyphonic character prediction model to obtain the pronunciation of each character in the content unit, including polyphonic characters, output by the polyphonic character prediction model according to the context of the content unit. Return to step 505 until all content units have been processed and then execute step 517.

[0102] Step 517: Combine the pronunciation results of all words obtained from all content units.

[0103] Then, based on the rhythmic tags carried in the interactive content and the pronunciation of each character, including polyphonic characters, the interactive content can be played to the user in a rhythmic voice manner, thereby achieving accurate human-computer interaction with polyphonic characters.

[0104] This specification provides an embodiment of a device for predicting the pronunciation of polyphonic characters, see [link to embodiment]. Figure 6 The device includes:

[0105] The interactive content acquisition module 601 is configured to obtain the interactive content that needs to be played to the user;

[0106] The prosody processing module 602 is configured to input the interactive content into a pre-trained prosody prediction model, and the prosody prediction model outputs interactive content with prosody tags; wherein, the prosody tags are used to indicate pauses during pronunciation.

[0107] The pronunciation processing module 603 is configured to detect whether each content unit segmented by each prosody tag in the interactive content contains a polyphonic character. If it does, the pronunciation of the polyphonic character in the content unit is obtained based on the context of the content unit.

[0108] In one embodiment of the device described in this specification, the pronunciation processing module 603 is configured to perform: inputting a content unit into a pre-trained polyphonic character prediction model to obtain the pronunciation of each character in the content unit, including polyphonic characters, output by the polyphonic character prediction model according to the context of the content unit.

[0109] In one embodiment of the device described in this specification, the pronunciation processing module 603 is configured to perform the following: after detecting that the content unit does not contain polyphonic characters, obtain the pronunciation of each character in the content unit according to a pre-set monophonic character dictionary; wherein the monophonic character dictionary includes the pronunciation of each monophonic character.

[0110] In one embodiment of the device described in this specification, a voice playback module (not shown in the figure) is further included;

[0111] The voice playback module is configured to perform the following: play the interactive content to the user in a rhythmic voice manner, based on the prosody tags carried in the interactive content and the pronunciation of each character, including polyphonic characters.

[0112] In one embodiment of the device described in this specification, the pronunciation processing module 603 is configured to perform the following: after detecting that a content unit contains only one character and that character is a polyphonic character, merge the content unit containing the polyphonic character with the adjacent content units into one content unit; input the merged content unit into a pre-trained polyphonic character prediction model to obtain the pronunciation of each character, including the polyphonic character, in the merged content unit, output by the polyphonic character prediction model according to the context of the merged content unit.

[0113] In one embodiment of the device described herein, a prosody prediction model training module (not shown in the figures) is further included, wherein prosody prediction model training is configured to perform:

[0114] Obtain the text corpus from the target business scenario;

[0115] Normalize the text corpus;

[0116] A sample corpus is obtained; wherein, the sample corpus is obtained by adding prosodic tags to the normalized text corpus; the prosodic tags are used to indicate the duration of pauses;

[0117] The model structure and parameters of the prosody prediction model are initialized using the model structure and parameters of the trained punctuation prediction model.

[0118] The sample corpus with prosodic labels is input into the initialized prosodic prediction model to train the prosodic prediction model.

[0119] In embodiments of the device described in this specification, the prosodic tags include at least one of the following: tags indicating word segmentation, tags indicating short pauses, tags indicating long pauses, and tags indicating the end of a sentence.

[0120] It should be noted that the above-mentioned devices are typically implemented on the server side. They can be set up on independent servers, or some or all of the devices can be combined and installed on the same server. This server can be a single server or a server cluster consisting of multiple servers. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system. The above-mentioned devices can also be implemented on computer terminals with strong computing capabilities.

[0121] This specification provides, in one embodiment, a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods of any embodiment in the specification.

[0122] This specification provides a computing device according to one embodiment, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to perform the method of any embodiment of the specification.

[0123] It is understood that the structures illustrated in the embodiments of this specification do not constitute a specific limitation on the apparatus of the embodiments of this specification. In other embodiments of the specification, the above-described apparatus may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0124] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0125] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, widgets, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0126] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment 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 on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the pronunciation of polyphonic characters, wherein the method includes: Obtain the interactive content that needs to be played to the user; The interactive content is input into a pre-trained prosodic prediction model, which then outputs interactive content with prosodic labels; the prosodic labels are used to indicate pauses during pronunciation. For each content unit segmented by each prosody tag in the interactive content, detect whether the content unit contains a polyphonic character. If it does, obtain the pronunciation of the polyphonic character in the content unit based on the context of the content unit. The training method for the prosody prediction model includes: Obtain the text corpus from the target business scenario; Normalize the text corpus; A sample corpus is obtained; wherein, the sample corpus is obtained by adding prosodic tags to the normalized text corpus; the prosodic tags are used to indicate the duration of pauses; The model structure and parameters of the prosody prediction model are initialized using the model structure and parameters of the trained punctuation prediction model. Input the sample corpus with prosodic labels into the initialized prosodic prediction model to train the prosodic prediction model. The initialization of the prosody prediction model's structure and parameters using the trained punctuation prediction model's structure and parameters further includes: removing fully connected layers from the punctuation prediction model's network structure; and setting the model's output category to the category included in the prosody label. The trained prosodic prediction model possesses the ability to predict punctuation marks, predict prosodic tags based on punctuation marks, and predict prosodic tags directly based on text content.

2. The method according to claim 1, wherein, The step of obtaining the pronunciation of the polyphonic character in the content unit based on the context of the content unit includes: inputting the content unit into a pre-trained polyphonic character prediction model to obtain the pronunciation of each character in the content unit, including the polyphonic character, output by the polyphonic character prediction model based on the context of the content unit.

3. The method according to claim 1, wherein, The method further includes: pre-setting a monosyllabic dictionary; the monosyllabic dictionary includes the pronunciation of each monosyllabic character; After detecting that the content unit does not contain polyphonic characters, it further includes: Based on a pre-set monosyllabic dictionary, the pronunciation of each character in the content unit is obtained.

4. The method according to claim 1, wherein, The method further includes: playing the interactive content to the user in a rhythmic speech manner, according to the prosody tags carried in the interactive content and the pronunciation of each character, including polyphonic characters.

5. The method according to claim 1, wherein, The method further includes: after detecting that a content unit contains only one character and that the character is a polyphonic character, before obtaining the pronunciation of the polyphonic character in the content unit, it further includes: merging the content unit containing the polyphonic character with the adjacent content units into one content unit; The step of obtaining the pronunciation of the polyphonic character included in the content unit based on the context of the content unit includes: inputting the merged content unit into a pre-trained polyphonic character prediction model to obtain the pronunciation of each character, including the polyphonic character, in the merged content unit, output by the polyphonic character prediction model based on the context of the merged content unit.

6. The method according to claim 1, wherein, The prosodic tags include at least one of the following: tags indicating word segmentation, tags indicating short pauses, tags indicating long pauses, and tags indicating the end of a sentence.

7. A device for predicting the pronunciation of polyphonic characters, wherein the device comprises: The interactive content acquisition module is configured to obtain the interactive content that needs to be played to the user; The prosody processing module is configured to input the interactive content into a pre-trained prosody prediction model, and the prosody prediction model outputs interactive content with prosody tags; wherein, the prosody tags are used to indicate pauses during pronunciation. The pronunciation processing module is configured to detect whether each content unit segmented by each prosody tag in the interactive content contains a polyphonic character. If it does, the module obtains the pronunciation of the polyphonic character in the content unit based on the context of the content unit. The device further includes a prosody prediction model training module, which is configured to perform: Obtain the text corpus from the target business scenario; Normalize the text corpus; A sample corpus is obtained; wherein, the sample corpus is obtained by adding prosodic tags to the normalized text corpus; the prosodic tags are used to indicate the duration of pauses; The model structure and parameters of the prosody prediction model are initialized using the model structure and parameters of the trained punctuation prediction model. Input the sample corpus with prosodic labels into the initialized prosodic prediction model to train the prosodic prediction model. The initialization of the prosody prediction model's structure and parameters using the trained punctuation prediction model's structure and parameters further includes: removing fully connected layers from the punctuation prediction model's network structure; and setting the model's output category to the category included in the prosody label. The trained prosodic prediction model possesses the ability to predict punctuation marks, predict prosodic tags based on punctuation marks, and predict prosodic tags directly based on text content.

8. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1-6.

9. A computing device comprising a memory and a processor, wherein the memory stores executable code, and the processor, when executing the executable code, implements the method of any one of claims 1-6.

Images