A long text-to-speech accuracy improvement method and device based on NLP

Abstract

This application relates to a method and apparatus for improving the accuracy of long text-to-speech conversion based on NLP. The method includes preprocessing the target text; performing multi-task semantic analysis on the preprocessed target text using a deep semantic parsing model to construct a long-distance semantic dependency graph and generate contextual labels; and generating the final speech signal of the target text based on the long-distance semantic dependency graph and contextual labels. This application proposes a three-level architecture of hierarchical preprocessing, dynamic semantic parsing, and adaptive prosodic generation. By integrating three core modules—an intelligent preprocessing module, a deep semantic parsing model, and an adaptive prosodic generator—it deeply integrates natural language processing and speech synthesis to achieve accurate speech conversion of complex text.

Description

Technical Field

[0001] This application relates to the field of long text-to-speech technology, and in particular to a method and apparatus for improving the accuracy of long text-to-speech based on NLP. Background Technology

[0002] Long text-to-speech technology is used in customer service for automated responses, in education for reading long texts aloud, and in content creation for audiobook generation, all requiring extremely high accuracy and naturalness of speech. Patents from companies like Zhongfu Safety Technology indicate that while existing technologies have improved processing speed and naturalness through algorithmic optimization, complex text processing still faces the following bottlenecks: Insufficient semantic understanding and processing capabilities for complex texts: Specifically, this manifests as biases in semantic parsing of polysemous words and technical terms, failure in processing long-distance semantic dependencies across paragraphs, and a lack of dynamic adaptability and robustness of text preprocessing to non-standard content. The accuracy and naturalness of speech generation are insufficient: including broken sentences in the generation of long and difficult sentences, unreasonable stress distribution, inability of semantic-acoustic mapping to dynamically adjust pronunciation parameters according to context, and lack of multimodal fusion capability for text sentiment and scene information, resulting in a disconnect between generated speech and application scenarios. Summary of the Invention

[0003] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0004] The main objective of this disclosure is to propose a method and apparatus for improving the accuracy of long text-to-speech conversion based on NLP, which can solve the accuracy problem of complex text processing in long text-to-speech technology.

[0005] Firstly, a method for improving the accuracy of long text-to-speech conversion based on NLP is provided, the method comprising: Preprocess the target text; A deep semantic parsing model is used to perform multi-task semantic analysis on the preprocessed target text, construct a long-distance semantic dependency graph and generate contextual labels. The final speech signal of the target text is generated based on long-distance semantic dependency graphs and contextual labels.

[0006] In some embodiments, the preprocessing of the target text includes: Remove illegal characters and formatting errors from the target text using regular expressions and word segmentation techniques; The semantic similarity between adjacent text segments is calculated based on the BERT model, and the text is segmented according to semantic boundaries; The similarity of terms is calculated based on the formula for matching initials and finals, and phonetic annotation is implemented.

[0007] In some embodiments, the formula for calculating term similarity based on the initial and final matching degree formula is as follows: ; in, For similarity, The length of the non-contiguous matching string. The length of the initial and final consonant string to be matched. Indicates the length of the initial and final vowel string of words in the domain dictionary.

[0008] In some embodiments, generating contextual tags includes: Contextual labels for the target text are generated based on dynamic Bayesian networks.

[0009] In some embodiments, generating the final speech signal of the target text includes: An attention-based Encoder-Decoder architecture captures the semantic information of the target text; Based on the captured semantic information, language features are generated; Based on the language features, the final speech signal of the target text is generated.

[0010] In some embodiments, generating the final speech signal of the target text based on the language features includes: Based on the aforementioned language features, generate the language signal of the target text; Dynamic planning is performed on the phoneme splicing in the speech signal to obtain the final speech signal.

[0011] In some embodiments, the dynamic planning of phoneme splicing in the speech signal to obtain the final speech signal includes: The phoneme splicing in the speech signal is dynamically planned based on the dynamic time warping algorithm to obtain the final speech signal.

[0012] Secondly, an NLP-based device for improving the accuracy of long text-to-speech conversion is provided, the device comprising: The data response module is used to respond to the text-to-image generation command and obtain preset prompt words; The reasoning module is used to set a consistency model and perform multi-step reasoning on the preset prompt words based on the consistency model. The strategy generation module is used to model the multi-step reasoning of the consistency model as a Markov decision process, and determine the optimal generation strategy of the consistency model based on the Markov decision process. The text graph generation module is used to model the reasoning process of the consistency model as a Markov decision process, and generate a text graph based on the noise and a preset reward function through the Markov decision process.

[0013] Thirdly, an electronic device is provided, including at least one controller and a memory for communicatively connecting with the controller; the memory stores instructions executable by the at least one controller, the instructions being executed by the at least one controller to cause the at least one controller to perform an NLP-based method for improving the accuracy of long text-to-speech conversion as described above.

[0014] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described method for improving the accuracy of long text-to-speech conversion based on NLP.

[0015] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for improving the accuracy of long text-to-speech based on NLP, provided in an embodiment of this application. Figure 2 This is a flowchart illustrating another embodiment of an NLP-based method for improving the accuracy of long text-to-speech conversion. Figure 3 This is a schematic diagram of the structure of a long text-to-speech accuracy improvement device based on NLP provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0020] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or function in a specific orientation, and therefore should not be construed as a limitation of this application.

[0021] like Figure 1 This application provides a method for improving the accuracy of long text-to-speech conversion based on NLP, the method including the following steps S100 to S300: Step S100: Preprocess the target text; Step S200: Perform multi-task semantic analysis on the preprocessed target text using a deep semantic parsing model to construct a long-distance semantic dependency graph and generate contextual labels; Step S300: Based on the long-distance semantic dependency graph and contextual tags, generate the final speech signal of the target text.

[0022] This method proposes a three-level architecture of hierarchical preprocessing, dynamic semantic parsing, and adaptive prosodic generation. By integrating three core modules—intelligent preprocessing module, deep semantic parsing model, and adaptive prosodic generator—it deeply integrates natural language processing and speech synthesis to achieve accurate speech conversion of complex text.

[0023] In some embodiments, the target text is preprocessed, including: Remove illegal characters and formatting errors from the target text using regular expressions and word segmentation techniques; The semantic similarity between adjacent text segments is calculated based on the BERT model, and the text is segmented according to semantic boundaries; The similarity of terms is calculated based on the formula for matching initials and finals, and phonetic annotation is implemented.

[0024] In some embodiments, the formula for calculating term similarity based on the initial and final matching degree formula is as follows: ; in, For similarity, The length of the non-contiguous matching string. The length of the initial and final consonant string to be matched. Indicates the length of the initial and final vowel string of words in the domain dictionary.

[0025] In some embodiments, generating contextual tags includes: Contextual labels for the target text are generated based on dynamic Bayesian networks.

[0026] In some embodiments, the final speech signal that generates the target text includes: An attention-based Encoder-Decoder architecture captures the semantic information of the target text; Based on the captured semantic information, language features are generated; Based on linguistic features, the final speech signal of the target text is generated.

[0027] In some embodiments, the final speech signal of the target text is generated based on language features, including: Generate the language signals of the target text based on language features; Dynamic programming is used to concatenate phonemes in the speech signal to obtain the final speech signal.

[0028] In some embodiments, dynamic planning is performed on the phoneme splicing in the speech signal to obtain the final speech signal, including: The phoneme splicing in the speech signal is dynamically planned based on the dynamic time warping algorithm to obtain the final speech signal.

[0029] For ease of understanding, one embodiment of this application provides a method for improving the accuracy of long text-to-speech conversion based on NLP, including the following steps: Step S910: Input text preprocessing flow; The preprocessing process is fundamental to ensuring the quality of subsequent semantic parsing and speech synthesis. This process comprises four key steps: text cleaning, semantic segmentation, domain dictionary matching, and standardized text sequence generation. These steps effectively resolve fundamental issues such as garbled text and terminological ambiguity, providing high-quality input for subsequent parsing.

[0030] (1) Text Cleaning Unit: Responsible for removing noise and redundant information from the input text to ensure its standardization and readability. Specific implementation includes removing garbled characters, standardizing formats, and handling special characters. The key to text cleaning lies in accurately identifying and processing abnormal characters that affect the quality of speech synthesis. For example, for processing numbers, consecutive digits need to be split into individual characters so that the subsequent speech synthesis system can pronounce them correctly. For Chinese text, the jieba library can be used for word segmentation to improve the accuracy of subsequent semantic segmentation.

[0031] (2) Semantic Segmentation Unit: Responsible for segmenting long texts into semantically coherent paragraphs or sentences, facilitating processing by the speech synthesis system. BERT-based semantic segmentation models can capture deep semantic information of text, achieving more accurate segmentation. The core of semantic segmentation is identifying semantic boundaries in text. BERT-based semantic segmentation models determine the optimal segmentation point by calculating the semantic similarity between adjacent text segments. When the similarity is below a threshold or reaches the maximum paragraph length, the system will perform segmentation to ensure the semantic coherence and processability of each paragraph.

[0032] (3) Domain dictionary loading unit: Responsible for resolving terminological ambiguity and improving the accuracy of speech synthesis. The text correction method based on initials and finals in speech recognition can effectively identify and correct pronunciation problems of domain terms. The core of domain dictionary matching is calculating the matching degree between the initials and finals sequence to be matched and the initials and finals sequences of words in the domain dictionary. The formula for calculating the matching degree is: ; Where m represents the length of the non-contiguous matching strings, d represents the length of the initial and final vowel string to be matched, and n represents the length of the initial and final vowel string of the word in the domain dictionary. This formula can effectively quantify the similarity between text fragments and domain terms, helping the system to accurately identify and correct terminological ambiguities.

[0033] Step S920, semantic-acoustic mapping; In step S9210, the attention-based Encoder-Decoder architecture effectively captures the semantic information of the text and generates high-quality speech features. The dynamic programming algorithm is then used to optimize phoneme concatenation and reduce the pronunciation error rate.

[0034] The Encoder-Decoder architecture based on attention mechanisms is the core of a TTS system. It encodes text sequences into semantic vectors and combines these vectors with contextual tags to generate acoustic feature parameters. In the FastSpeech 2 model, this architecture comprises three main modules: an encoder, a speech feature information predictor, and a decoder. The encoder module uses the Transformer's self-attention mechanism to encode text sequences into semantic vectors. The calculation formula for the self-attention mechanism is as follows: ; Where Q, K, and V represent the query, key, and value vectors, respectively. This represents the dimension of the key vector.

[0035] Multi-head attention mechanisms further extend self-attention mechanisms by computing multiple attention heads in parallel to capture different features of the text. ; The formula for calculating each attention point is: ; in, , and It is a learnable parameter matrix used to map the input. It is the output transformation matrix.

[0036] Step S9220: Dynamic programming optimizes phoneme splicing; Dynamic programming algorithms are used to optimize the phoneme concatenation process and reduce pronunciation error rates. In speech synthesis, phoneme concatenation needs to consider the transition relationships between phonemes and pronunciation accuracy. Dynamic programming ensures the naturalness and accuracy of phoneme concatenation by finding the optimal path.

[0037] DTW (Dynamic Time Warping) is a commonly used dynamic programming method for comparing the similarity of two time series. In phoneme concatenation, DTW can be used to calculate the optimal alignment path between the phoneme sequence and the acoustic features.

[0038] The mathematical expression for the DTW algorithm is:

[0039] Where Dist(i,j) represents the distance between phoneme sequence i and acoustic feature j, which can be Euclidean distance or other similarity measures.

[0040] Dynamic programming algorithms optimize the phoneme splicing process by calculating the optimal alignment path between phoneme sequences and acoustic features, thereby improving the accuracy and naturalness of speech synthesis.

[0041] This embodiment employs an end-to-end training strategy, which is crucial for long text-to-speech systems. By using a hybrid loss function (WER + MOS) and transfer learning, the model's generalization ability in complex text scenarios is improved. The hybrid loss function can simultaneously optimize the accuracy and naturalness of speech synthesis, while transfer learning utilizes knowledge from the source domain (such as general corpora) to enhance the model's performance in the target domain (such as customer service dialogues or long text reading).

[0042] Hybrid Loss Function Design: The hybrid loss function combines two evaluation metrics: Word Error Rate (WER) and Naturalness Score (MOS), optimizing model performance through linear weighting. WER measures the accuracy of speech synthesis, while MOS measures the naturalness of speech; combining the two provides a comprehensive evaluation of the model's performance.

[0043] The formula for calculating WER is: ; Where S represents the number of replacement errors, D represents the number of deletion errors, I represents the number of insertion errors, and N represents the total number of words in the reference text, 26.

[0044] MOS scoring typically employs a regression task, using mean squared error (MSE) as the loss function: ; in, This represents the true MOS score of the i-th sample. This represents the model's predicted value.

[0045] Transfer learning strategies: Transfer learning strategies leverage the knowledge of pre-trained models in the source domain (such as general corpora) to improve the model's performance in the target domain (such as customer service dialogues or long text reading). By using techniques such as freezing some layers, adding domain-adaptive layers, and data augmentation, it is possible to effectively fine-tune the model using limited target domain data.

[0046] The core of transfer learning strategies is to leverage the knowledge learned by a pre-trained model in the source domain to reduce the need for data in the target domain. By freezing the first few layers of the pre-trained model, the general semantic representations learned in the source domain can be preserved, while adding domain adaptation layers allows the model to be adjusted to suit the characteristics of the target domain. This strategy is particularly suitable for resource-constrained scenarios, such as customer service dialogues or long text readings in specialized fields.

[0047] like Figure 3 One embodiment of this application provides an NLP-based device for improving the accuracy of long text-to-speech conversion. The device includes: The preprocessing module 1001 is used to preprocess the target text; The semantic analysis module 1002 is used to perform multi-task semantic analysis on the preprocessed target text through a deep semantic parsing model, construct a long-distance semantic dependency graph and generate contextual labels; The speech conversion module 1003 is used to generate the final speech signal of the target text based on long-distance semantic dependency graph and contextual labels.

[0048] It should be noted that the NLP-based long text-to-speech accuracy improvement device provided in this embodiment is based on the same inventive concept as the aforementioned NLP-based long text-to-speech accuracy improvement method. Therefore, the relevant content of the aforementioned NLP-based long text-to-speech accuracy improvement method also applies to the content of the NLP-based long text-to-speech accuracy improvement device, and therefore will not be repeated here.

[0049] like Figure 4 This application also provides an electronic device, which includes: At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the above-described method for improving the accuracy of long text-to-speech based on NLP.

[0050] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0051] The electronic device according to embodiments of this application will now be described in detail. The electronic device includes: The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to execute a method for improving the accuracy of long text-to-speech based on NLP, according to an embodiment of this disclosure.

[0052] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 communicate with each other within the device via bus 2000.

[0053] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described method for improving the accuracy of NLP-based long text-to-speech conversion.

[0054] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, which can be linked to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0055] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.

[0056] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0058] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0059] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0060] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication links shown or discussed may be through some interfaces; the indirect coupling or communication links between apparatuses or units may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

Claims

1. A method for improving the accuracy of long text-to-speech conversion based on NLP, characterized in that, The method includes: Preprocess the target text; A deep semantic parsing model is used to perform multi-task semantic analysis on the preprocessed target text, construct a long-distance semantic dependency graph and generate contextual labels. Based on long-distance semantic dependency graphs and contextual labels, the final speech signal of the target text is generated; The preprocessing of the target text includes: Remove illegal characters and formatting errors from the target text using regular expressions and word segmentation techniques; The semantic similarity between adjacent text segments is calculated based on the BERT model, and the text is segmented according to semantic boundaries; The similarity of terms is calculated based on the initial and final matching formula to achieve phonetic transcription; the formula for calculating the similarity of terms based on the initial and final matching formula is as follows: ； in, For similarity, The length of the non-contiguous matching string. The length of the initial and final consonant string to be matched. Indicates the length of the initial and final vowel string of words in the domain dictionary.

2. The method for improving the accuracy of long text-to-speech based on NLP according to claim 1, characterized in that, The generated contextual tags include: Contextual labels for the target text are generated based on dynamic Bayesian networks.

3. The method for improving the accuracy of long text-to-speech based on NLP according to claim 2, characterized in that, The final speech signal used to generate the target text includes: An attention-based Encoder-Decoder architecture captures the semantic information of the target text; Based on the captured semantic information, language features are generated; Based on the language features, the final speech signal of the target text is generated.

4. The method for improving the accuracy of long text-to-speech based on NLP according to claim 3, characterized in that, The step of generating the final speech signal of the target text based on the language features includes: Based on the aforementioned language features, generate the language signal of the target text; Dynamic planning is performed on the phoneme splicing in the speech signal to obtain the final speech signal.

5. The method for improving the accuracy of long text-to-speech based on NLP according to claim 4, characterized in that, The dynamic planning of phoneme splicing in the speech signal to obtain the final speech signal includes: The phoneme splicing in the speech signal is dynamically planned based on the dynamic time warping algorithm to obtain the final speech signal.

6. A device for improving the accuracy of long text-to-speech conversion based on NLP, characterized in that, The device includes: The preprocessing module is used to preprocess the target text; The semantic analysis module is used to perform multi-task semantic analysis on the preprocessed target text through a deep semantic parsing model, construct a long-distance semantic dependency graph and generate contextual labels. The speech conversion module is used to generate the final speech signal of the target text based on a long-distance semantic dependency graph and contextual tags; the preprocessing of the target text includes: Remove illegal characters and formatting errors from the target text using regular expressions and word segmentation techniques; The semantic similarity between adjacent text segments is calculated based on the BERT model, and the text is segmented according to semantic boundaries; The similarity of terms is calculated based on the initial and final matching formula to achieve phonetic transcription; the formula for calculating the similarity of terms based on the initial and final matching formula is as follows: ； in, For similarity, The length of the non-contiguous matching string. The length of the initial and final consonant string to be matched. Indicates the length of the initial and final vowel string of words in the domain dictionary.

7. An electronic device, characterized in that, It includes at least one controller and a memory for communicatively connecting with the controller; the memory stores instructions executable by the at least one controller, which, when executed by the at least one controller, causes the at least one controller to perform an NLP-based method for improving the accuracy of long text-to-speech conversion as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform an NLP-based method for improving the accuracy of long text-to-speech as described in any one of claims 1 to 5.

Images