SSML tag processing method and device for speech synthesis, equipment and storage medium

Abstract

This application discloses a method, apparatus, device, and storage medium for SSML tag processing in speech synthesis. The method includes: acquiring and parsing the text to be synthesized to obtain semantic parsing results, including heterogeneous date entities and custom SSML tags; converting the heterogeneous date entities according to a preset neutral standard date structure and converting the custom SSML tags into neutral abstract tags; acquiring the configuration mapping rules corresponding to the target speech synthesis engine, mapping the neutral standard date entities to the target text to be read, and mapping the neutral abstract tags to private SSML tags; concatenating the target text to be read and the private SSML tags to generate target-specific SSML tags to guide the target speech synthesis engine in speech synthesis. This application achieves the conversion from neutral semantics to engine-private syntax through configuration mapping rules, eliminating the need to modify business logic when switching or adding speech synthesis engines, thereby improving the system's scalability.

Description

Technical Field

[0001] This application relates to the field of speech synthesis technology, and in particular to an SSML tag processing method, apparatus, device and storage medium for speech synthesis. Background Technology

[0002] Text-to-speech (TTS) technology has been widely applied in scenarios such as intelligent customer service, content broadcasting, and in-vehicle voice interaction. In practical engineering scenarios, business systems face two types of adaptation problems: First, date and time texts exhibit heterogeneous forms such as multiple regional word orders, multiple delimiters, multiple calendar systems, and colloquial expressions. The same date in different formats may lead to errors in reading or disordered word order by the TTS engine. Second, each speech synthesis engine vendor provides proprietary extended Speech Synthesis Markup Language (SSML) tags to control the synthesis effects of the TTS engine, such as pauses, speech rate, and intonation. The SSML tag syntax of different vendors is incompatible.

[0003] To address the aforementioned issues, existing technologies primarily employ an interface-layer encapsulation approach. This involves building a unified calling interface between the business system and the TTS engine, adapting the parameter formats of the Application Programming Interface (API) to achieve tag adaptation and date text standardization. However, this solution only masks vendor differences at the interface call level and does not provide a unified model for date text formats and SSML tag rules at the semantic layer. Date format conversion and tag adaptation are still written separately by the business side according to vendor specifications, resulting in large-scale modifications to business code when switching TTS engines, leading to high adaptation costs and limited scalability. Summary of the Invention

[0004] The main purpose of this application is to provide an SSML tag processing method, apparatus, device and storage medium for speech synthesis, which aims to solve the technical problems of existing tag adaptation and date text standardization relying on the interface call layer to shield vendor differences, resulting in large-scale modification of business code when switching TTS engines, high adaptation costs and limited scalability.

[0005] To achieve the above objectives, this application proposes an SSML tag processing method for speech synthesis, which includes: Obtain the text to be synthesized into speech, and parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags. The heterogeneous date entities are converted according to a preset neutral standard date structure, and the custom SSML tags are converted into neutral abstract tags; The corresponding configuration mapping rules are obtained according to the target speech synthesis engine, and the neutral standard date entity obtained by conversion is mapped to the target reading text based on the configuration mapping rules, and the neutral abstract tag is mapped to the private SSML tag corresponding to the target speech synthesis engine. The target text to be read aloud is concatenated with the private SSML tag to generate a target-specific SSML tag, which is used to guide the target speech synthesis engine to perform speech synthesis.

[0006] In one embodiment, the semantic parsing result further includes free text, and the step of converting the heterogeneous date entity according to a preset neutral standard date structure includes: The heterogeneous date entities are converted according to the preset neutral standard date structure to obtain several candidate neutral standard conversion entities; Based on the free text, each of the candidate neutral standard transformation entities is scored according to a preset scoring dimension, which includes at least two of the following: regional custom dimension, contextual semantic dimension, and timeliness dimension. Based on the scoring results, the candidate neutral standard conversion entity with the highest score is determined as the neutral standard date entity.

[0007] In one embodiment, the step of scoring each of the candidate neutral standard transformation entities based on the free text according to a preset scoring dimension includes: Based on the regional configuration information of the target speech synthesis engine, determine the regional habit score of each of the candidate neutral standard conversion entities; Based on the free text, determine the context words around the heterogeneous date entity in the text to be synthesized, and determine the context semantic score of each of the candidate neutral standard transformation entities according to the context words; The timeliness score of each candidate neutral standard conversion entity is determined based on the time distance between the current date and the corresponding date of each candidate neutral standard conversion entity.

[0008] In one embodiment, the step of converting the custom SSML tag into a neutral abstract tag includes: Extract the corresponding factor parameters from the custom SSML tags according to the preset neutral label factors. The preset label factors include: sentiment sub-tags, speech rate sub-tags, pause sub-tags, intonation sub-tags, and sentence break tags. By integrating the parameters of each factor, the neutral abstract label is obtained.

[0009] In one embodiment, the steps of obtaining the corresponding configuration mapping rules according to the target speech synthesis engine, mapping the converted neutral standard date entities to the target read-aloud text based on the configuration mapping rules, and mapping the neutral abstract tags to the private SSML tags corresponding to the target speech synthesis engine include: In the rule configuration library, the configuration mapping rules corresponding to the target speech synthesis engine are matched. The rule configuration library contains pre-stored configuration mapping rules between the engine specification language structure and the preset neutral standard date structure corresponding to different speech synthesis engines. The target date mapping rule and the target tag mapping rule are determined based on the configured mapping rules. Based on the target date mapping rule, the neutral standard date entity is mapped to the target read-aloud text, and based on the target tag mapping rule, the neutral abstract tag is mapped to the private SSML tag corresponding to the target speech synthesis engine.

[0010] In one embodiment, the step of concatenating the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag includes: The target splicing method is determined according to the configured mapping rules; The target text to be read aloud and the private SSML tag are concatenated based on the target concatenation method to obtain the concatenation result; Add semantic metadata annotations to the concatenation result to generate target-specific SSML tags; The semantic metadata annotation includes a conversion link representation from the heterogeneous date entity to the target read-aloud text and a conversion link representation from the custom SSML tag to the private SSML tag.

[0011] In one embodiment, after the step of obtaining the corresponding configuration mapping rules based on the target speech synthesis engine, the method further includes: Determine if there are any instances of target text conversion failure or private SSML tag conversion failure; If so, a preset downgrade strategy is executed to generate standard compatible SSML tags, which are used to guide the target speech synthesis engine to perform speech synthesis. The preset degradation strategy includes: outputting the neutral standard date entity as the original text format or outputting the remaining text after removing the neutral abstract tags that cannot be mapped.

[0012] Furthermore, to achieve the above objectives, this application also proposes an SSML tagging processing device for speech synthesis, the device comprising: The text parsing module is used to acquire the text to be synthesized into speech and to parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags. The normalization conversion module is used to convert the heterogeneous date entities according to a preset neutral standard date structure and convert the custom SSML tags into neutral abstract tags; The rule mapping module is used to obtain the corresponding configuration mapping rules according to the target speech synthesis engine, and to map the converted neutral standard date entity to the target reading text based on the configuration mapping rules, and to map the neutral abstract tag to the private SSML tag corresponding to the target speech synthesis engine. The rendering output module is used to concatenate the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag. The target-specific SSML tag is used to guide the target speech synthesis engine to perform speech synthesis.

[0013] Furthermore, to achieve the above objectives, this application also proposes an SSML tag processing device for speech synthesis, the device comprising: a memory, a processor, and an SSML tag processing program for speech synthesis stored in the memory and executable on the processor, the SSML tag processing program for speech synthesis being configured to implement the steps of the SSML tag processing method for speech synthesis as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, storing an SSML tag processing program for speech synthesis, wherein when the SSML tag processing program for speech synthesis is executed by a processor, it implements the steps of the SSML tag processing method for speech synthesis as described above.

[0015] This application discloses an SSML tag processing method for speech synthesis. The method includes: acquiring the text to be synthesized and parsing it to obtain a semantic parsing result, the semantic parsing result including at least: heterogeneous date entities and custom SSML tags; converting the heterogeneous date entities according to a preset neutral standard date structure and converting the custom SSML tags into neutral abstract tags; acquiring the corresponding configuration mapping rules according to the target speech synthesis engine, and mapping the converted neutral standard date entities to the target reading text based on the configuration mapping rules, and mapping the neutral abstract tags to the private SSML tags corresponding to the target speech synthesis engine; concatenating the target reading text and the private SSML tags to generate target-specific SSML tags, the target-specific SSML tags being used to guide the target speech synthesis engine to perform speech synthesis.

[0016] This application achieves dual decoupling between date text format and SSML tags and specific speech synthesis engines by uniformly converting heterogeneous date entities into neutral standard date structures and custom SSML tags into neutral abstract tags, and then adapting them to different target speech synthesis engines through configuration mapping rules. The business side only needs to maintain one set of neutral semantic expressions, eliminating the need to write multiple sets of business code for different vendors, thus reducing adaptation costs. Furthermore, this application completes the conversion from neutral semantics to engine-specific syntax through configuration mapping rules, so that switching or adding a speech synthesis engine only requires updating the mapping rule configuration without modifying the business logic, improving the system's scalability and compatibility. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the SSML tag processing method for speech synthesis according to this application. Figure 2 This is a flowchart illustrating the second embodiment of the SSML tag processing method for speech synthesis in this application. Figure 3 This is a flowchart illustrating the third embodiment of the SSML tag processing method for speech synthesis in this application. Figure 4 This is a schematic diagram of the entire process of the SSML tag processing method for speech synthesis in this application. Figure 5 The device decoupling architecture block diagram for implementing the SSML tag processing method for speech synthesis in this application is shown below; Figure 6 This is a schematic diagram of the module structure of the SSML tag processing device for speech synthesis according to this application; Figure 7 This is a schematic diagram of the SSML tag processing device for speech synthesis according to this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] This application provides an SSML tagging method for speech synthesis, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the SSML tag processing method for speech synthesis according to this application. In this embodiment, the method includes steps S10 to S40: Step S10: Obtain the text to be synthesized into speech and parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags.

[0024] It should be noted that the method of this embodiment can be applied to scenarios requiring text-to-speech (TTS), such as intelligent customer service, news broadcasting, and in-vehicle voice interaction. The executing entity of this embodiment can be a computing electronic device with functions such as data processing, network communication, and program execution, such as a desktop computer, server, or edge computing device. Here, a speech synthesis preprocessing device (referred to as "device") is used as an example to illustrate this embodiment and the following embodiments.

[0025] It should be understood that the text to be synthesized can be the raw text directly input by the business system. This text can contain free text content that needs to be read aloud, as well as date-type text that needs to be specially processed during speech synthesis and markup language tags that need to control the speech synthesis effect.

[0026] Heterogeneous date entities refer to date texts with different regional word order, different separators, different year reckoning methods, or colloquial expressions. For example, a heterogeneous date entity can be represented as a hyphenated format like "2026-04-28", as a slash-separated format like "4 / 28 / 2026", as a relative date expression like "next Wednesday", or as a traditional year reckoning like "the tenth day of the third lunar month".

[0027] Custom SSML tags can be markup language tags that business systems can customize to control the speech synthesis effect, such as tags for controlling pauses, tags for controlling speech rate, or tags for controlling intonation.

[0028] In a specific implementation, after the device obtains the text to be speech-synthesized, it can scan the text through regular expression matching or a semantic parsing model to identify date-type text segments and custom SSML tag segments therein, distinguish them from ordinary free text, and form a structured semantic parsing result, that is, obtain heterogeneous date entities and custom SSML tags.

[0029] Step S20: Convert the heterogeneous date entities according to a preset neutral standard date structure, and convert the custom SSML tags into neutral abstract tags.

[0030] It should be noted that the preset neutral standard date structure can be a unified date semantic expression format that is pre-set and independent of a specific speech synthesis engine. The preset neutral standard date structure can include fields such as year, month, day, hour, minute, time zone, and time granularity, and is used to carry the standardized semantic information after parsing of heterogeneous date entities.

[0031] It should be understood that after the heterogeneous date entities are converted, they no longer retain their original writing format or word order, but are uniformly expressed as this neutral standard date structure.

[0032] For example, no matter whether the input heterogeneous date entity is "2026-04-28", "4 / 28 / 2026", or "April 28, 2026", after conversion, a structured semantic object containing year = 2026, month = 4, and day = 28 is output.

[0033] The process of converting custom SSML tags into neutral abstract tags is to convert tags defined by the business system and independent of specific manufacturers into a standardized, engine-independent neutral tag expression. The neutral abstract tag can include semantic information such as emotion type, speech rate value, pause duration, intonation type, and sentence-breaking position, but does not include any manufacturer-specific grammar structures.

[0034] In a specific implementation, the device can call corresponding parsers to convert different types of heterogeneous date entities into a preset neutral standard date structure; at the same time, for different types of custom SSML tags, extract their semantic parameters and encapsulate them into neutral abstract tags.

[0035] Step S30: Obtain the corresponding configuration mapping rule according to the target speech synthesis engine, and map the converted neutral standard date entity into the target reading text based on the configuration mapping rule, and map the neutral abstract tag into the private SSML tag corresponding to the target speech synthesis engine.

[0036] It should be noted that the target speech synthesis engine (target TTS engine) can be a specific engine that actually performs speech synthesis, such as the TTS service of vendor A or vendor B. Different speech synthesis engines have different proprietary SSML tag syntaxes, and their requirements for the reading format of date text may also differ.

[0037] Configuration mapping rules are pre-configured sets of rules that describe the correspondence between neutral standard date structures and target text formats, as well as between neutral abstract tags and target engine-specific SSML tags. These configuration mapping rules can be associated with the identifier of the target TTS engine; different target TTS engines can correspond to different configuration mapping rules.

[0038] After obtaining the converted neutral standard date entity, the device retrieves the corresponding configuration mapping rules from the rule base according to the currently selected target TTS engine, and organizes the year, month, day and other fields in the neutral standard date entity into the target text to be read aloud according to the format required by the target TTS engine.

[0039] For example, for a Chinese-language TTS engine, the neutral standard date entity {year=2026, month=4, day=28} can be mapped to "April 28, 2026"; for an American English-language TTS engine, it can be mapped to "April 28, 2026".

[0040] At the same time, the device converts the semantic information in the neutral abstract tags into a private SSML tag syntax that the target TTS engine can recognize, according to the same configuration mapping rules.

[0041] For example, a neutral pause label can be mapped to vendor A's <break time="500ms" / > , or mapped to vendor B's {{pause:500}}.

[0042] Step S40: Concatenate the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag. The target-specific SSML tag is used to guide the target speech synthesis engine to perform speech synthesis.

[0043] It should be noted that after mapping the date entity and SSML tag respectively, the device can concatenate the mapped target text and private SSML tag with the original free text fragment in the order of the original text to form a complete target-specific SSML tag that can be directly sent to the target TTS engine.

[0044] The target-specific SSML tag no longer contains the original heterogeneous date entity or custom SSML tag; instead, it is entirely replaced with a format that the target speech synthesis engine can correctly parse and read aloud. Therefore, upon receiving the target-specific SSML tag, the target TTS engine can synthesize the expected speech according to the target text to be read aloud and the private SSML tag instructions.

[0045] Furthermore, considering that mapping anomalies may occur during the actual mapping conversion process, this embodiment can also introduce an anomaly tolerance mechanism. Therefore, after the aforementioned step of obtaining the corresponding configuration mapping rules based on the target speech synthesis engine, the following steps are also included: S501~S502: Step S501: Determine whether there is a failure to convert the target text or a failure to convert the private SSML tag.

[0046] It should be noted that during the actual mapping and conversion process, considering possible exceptions such as incomplete mapping rule configuration, the target speech synthesis engine not supporting a certain tag syntax, or the date format exceeding the preset range, the device can verify the conversion result after performing the mapping and conversion.

[0047] Step S502: If yes, then execute the preset downgrade strategy to generate standard compatible SSML tags, which are used to guide the target speech synthesis engine to perform speech synthesis.

[0048] It should be noted that standard-compliant SSML tags can be SSML text that does not contain any vendor-specific extension tags and only uses the most basic general syntax, or output degenerates into plain text. These standard-compliant SSML tags can be correctly parsed by common TTS engines, ensuring that the speech synthesis process will not be interrupted due to parsing failures.

[0049] The preset degradation strategy allows for different degradation processes to be executed depending on the failure scenario. For cases where date entity conversion fails, date format standardization degradation is performed, meaning that mapping the neutral standard date entity to the target text is abandoned, and the original text format of the date entity is directly output. For example, if the input date entity is "2026-04-28" and the mapping rule fails to match, then "2026-04-28" is output as the date content without any format conversion.

[0050] For cases where custom SSML tag conversion fails, tag removal and degradation can be performed. This involves removing unmapped neutral abstract tags from the synthesized data, retaining only the free text content for output. These two degradation strategies can be executed in parallel without affecting each other. For example, if the input is a custom pause tag " <cus-pause / >If a private pause tag cannot be mapped to the target engine, then the tag is removed from the synthetic data, and only the free text surrounding it is retained.

[0051] In addition, the preset degradation strategy also includes a minimum degradation strategy. The minimum degradation strategy means that the device discards all date conversion results and tag conversion results, outputs plain text format free text content, or outputs SSML text containing only the most basic general syntax, to ensure that the speech synthesis process can be executed with the lowest usable standard.

[0052] It should be understood that the device can execute corresponding degradation strategies based on different conversion failure scenarios to obtain a degraded date conversion result and / or a degraded tag conversion result. The degraded date conversion result is the target text after degradation processing, and the degraded tag conversion result is the private SSML tags after degradation processing.

[0053] If the aforementioned conversion failures still exist in the downgraded date conversion results or downgraded tag conversion results, the first step is to determine the specific number of failure items, including the number of date entities that failed to convert and the number of abstract tags that failed to convert. Specifically, the number of date entities that failed to convert refers to the number of neutral standard date entities that failed to be successfully mapped to the target text, and the number of abstract tags that failed to convert refers to the number of neutral abstract tags that failed to be successfully mapped to private SSML tags.

[0054] If the number of failed items is less than the total number of convertible items, the device returns to the downgrade strategy corresponding to different conversion failure scenarios until there are no conversion failures in the downgraded date conversion result and the downgraded tag conversion result. The total number of convertible items includes the total number of date entities in the neutral standard date entity and the total number of abstract tags in the neutral abstract tag. When there are no more conversion failures in the downgraded result, the device can return the obtained target text and private SSML tags to the aforementioned step S40 to concatenate them into the target-specific SSML tags.

[0055] If the number of failed items equals the total number of convertible items, or if the number of failed items still equals the total number of convertible items after repeatedly executing the degradation strategy, then the system can execute the minimum degradation strategy and generate standard compatible SSML tags.

[0056] For example, if the text to be synthesized contains three date entities and one pause label, and only one date entity fails to be mapped, after performing date format standardization downgrade, that date entity will output the original text, while the other two date entities and one pause label will be mapped successfully, then the downgrade is complete.

[0057] If the number of failed items is equal to the total number of convertible items, or if the number of failed items is still equal to the total number of convertible items after repeated downgrading, then the minimum downgrading strategy is executed to generate standard compatible SSML tags. For example, if the text to be synthesized contains a date entity and a pause tag, both of which fail to map and cannot be successfully mapped even after repeated downgrading, then the number of failed items (2) is equal to the total number of convertible items (2), and the minimum downgrading strategy is executed.

[0058] It should also be noted that the aforementioned preset degradation strategies and configuration mapping rules are stored together in the rule configuration library. Configuration mapping rules support independent configuration by four dimensions: language, region, target TTS engine, and business scenario. When a new adaptation object is added, the device only needs to add the corresponding configuration item in the rule configuration library; no modification to the system source code is required. During mapping conversion, the device dynamically loads the corresponding configuration mapping rules and degradation strategy parameters based on the current target speech synthesis engine, language configuration, region configuration, and business scenario identifier, and performs the corresponding mapping conversion and degradation processing.

[0059] This embodiment achieves decoupling of date text format and SSML tags from specific speech synthesis engines by converting heterogeneous date entities into neutral standard date structures, converting custom SSML tags into neutral abstract tags, and then performing reverse mapping based on the target speech synthesis engine's configuration mapping rules. The business system only needs to maintain one set of neutral semantic expressions, eliminating the need to write multiple sets of business code for different vendors. Simultaneously, a preset degradation strategy automatically generates standard-compatible SSML when mapping fails, ensuring the basic availability of the speech synthesis service.

[0060] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the SSML tag processing method for speech synthesis according to this application.

[0061] In this embodiment, to specifically illustrate how to perform neutralization conversion on heterogeneous date entities and custom SSML label generators, step S20 specifically includes: steps S201~S205: Step S201: Convert the heterogeneous date entity according to the preset neutral standard date structure to obtain a number of candidate neutral standard conversion entities.

[0062] It should be noted that the same heterogeneous date entity can have multiple parsing methods. For example, the input date text "04-05-2026" can be parsed as either "month=4, day=5, year=2026" or "month=5, day=4, year=2026". During the parsing phase, the device can convert all possible parsing results into corresponding neutral standard date structures, forming several candidate neutral standard conversion entities. Each candidate neutral standard conversion entity contains standardized fields such as year, month, and day, but the values ​​of these fields differ.

[0063] In practice, the device can enumerate all parsing candidates that meet the basic numerical constraints based on the type of date separator, the range of values ​​for each value (whether the month is between 1 and 12, whether the date is within a reasonable range), and common date format rules.

[0064] Step S202: Based on the free text, score each of the candidate neutral standard transformation entities according to preset scoring dimensions. The preset scoring dimensions include at least two of the following: regional custom dimension, contextual semantic dimension, and timeliness dimension.

[0065] It should be noted that the free text can be ordinary text content in the text to be synthesized, excluding heterogeneous date entities and custom SSML tags. The free text can contain date-related contextual clues, such as seasonal words, holiday names, and relative time indicators, which can be used to help determine the rationality of date parsing.

[0066] Preset scoring dimensions may include regional customs, contextual semantics, and timeliness. Devices can select at least two of these dimensions for scoring based on the actual application scenario, or they can use all three dimensions for comprehensive evaluation.

[0067] In addition, the scoring dimension selection strategy can be associated with the regional configuration of the target speech synthesis engine. For example, all three dimensions can be enabled when targeting cross-border business scenarios, while only the regional habit dimension and the timeliness dimension can be enabled when targeting a single regional scenario.

[0068] Furthermore, to specifically illustrate how to score each candidate neutral standard transformation entity, step S202 specifically includes: steps S2021~S2023: Step S2021: Determine the regional habit score of each of the candidate neutral standard conversion entities based on the regional configuration information of the target speech synthesis engine.

[0069] It should be understood that different regions have different conventions for writing date order. For example, American English typically uses "month / day / year," European English uses "day / month / year," and Chinese typically uses "year-month-day." The geographic configuration information of the target TTS engine can be used to indicate the target geographic region of the current speech synthesis service. For example, when the target geographic region is configured as the United States, it tends to select parsing results with the month first and the day last.

[0070] In its implementation, the device presets a base score for each date sequence based on the geographic configuration information. For example, in the US geographic configuration, the candidate score for the "month / day / year" sequence is positive, while the candidate score for the "day / month / year" sequence is negative or zero. The device assigns a corresponding geographic custom score based on the sequence type of each candidate.

[0071] Step S2022: Determine the context words around the heterogeneous date entity in the text to be synthesized based on the free text, and determine the context semantic score of each candidate neutral standard transformation entity according to the context words.

[0072] It should be noted that the contextual words surrounding date entities in the text to be synthesized may contain semantic information such as seasonal words (e.g., "spring," "autumn"), holiday names (e.g., "Christmas," "Spring Festival"), and relative time indicators (e.g., "next week," "yesterday"). This semantic information can provide additional basis for date parsing.

[0073] In its implementation, the device can extract text fragments within a preset range before and after heterogeneous date entities as a context window, and perform semantic analysis on the words within the window. For example, if "spring" appears in the context, it tends to select candidates for spring (such as March to May); if "Christmas" appears, it tends to select candidates for December. The device calculates a context semantic score based on the degree of matching between the corresponding date of each candidate and the context semantics.

[0074] Step S2023: Determine the timeliness score of each candidate neutral standard conversion entity based on the time distance between the current date and the corresponding date of each candidate neutral standard conversion entity.

[0075] It should be noted that in date resolution, if multiple reasonable candidates exist, the device may generally tend to select the date that is closer to the current date and has not yet expired. For example, if the current date is May 10, 2021, and there are two candidates for "04-05-2021" (April 5 and May 4), April 5 has expired, while May 4 has not. In this case, the timeliness score should favor the candidate that has not expired.

[0076] In practice, the device can obtain the current system date and calculate the time difference (in days) between each candidate date and the current date. Candidates pointing towards the future are given positive scores; candidates pointing towards the past are given negative scores based on the number of days exceeding the current date; and candidates closer to the current date receive higher timeliness scores.

[0077] Step S203: Based on the scoring results, the candidate neutral standard conversion entity with the highest score is determined as the neutral standard date entity.

[0078] It should be noted that after scoring each dimension, the device can perform a weighted summation of the scores for each candidate neutral standard transformation entity across all dimensions to obtain a comprehensive score for each candidate neutral standard transformation entity. The weighting coefficients can be dynamically configured according to the business scenario. For example, in a cross-border broadcasting scenario, the weight of the regional habits dimension can be increased, and in a news broadcasting scenario, the weight of the timeliness dimension can be increased.

[0079] Specifically, the device can select the candidate neutral standard transformation entity with the highest comprehensive score as the final neutral standard date entity output, and include the confidence information of this entity in the output result. This confidence level can be calculated based on the difference between the highest and second-highest scores, and is used to subsequently determine whether manual intervention is needed or a more stringent verification strategy should be implemented.

[0080] Step S204: Extract the corresponding factor parameters from the custom SSML tags according to the preset neutral label factors. The preset label factors include: sentiment sub-tags, speech rate sub-tags, pause sub-tags, intonation sub-tags, and sentence break tags.

[0081] It should be noted that custom SSML tags can be markers defined by the business system itself to control the speech synthesis effect. Different business systems may define different forms of custom tags. The device can uniformly convert the above heterogeneous custom tags into standardized neutral abstract tags so that they can be subsequently mapped to the proprietary SSML syntax of different manufacturers.

[0082] It should be understood that the preset neutral label factor can be a standardized classification of the control effect of speech synthesis, including: emotion sub-label, speech rate sub-label, pause sub-label, intonation sub-label, and sentence segmentation label.

[0083] Among them, the emotion sub-tag is used to identify the type of emotion to be expressed when reading aloud, such as happy, sad, serious, etc.; the speech rate sub-tag is used to identify the reading speed, such as fast, normal, slow; the pause sub-tag is used to identify the pause position and pause duration; the intonation sub-tag is used to identify intonation changes, such as rising, falling; and the sentence break tag is used to identify the sentence break position within a sentence.

[0084] In practical implementation, the device can define corresponding extraction rules for each preset label factor. For example, for the custom pause label " <pause duration="500ms" / > The device extracts the following parameters: "pause sub-tag, duration 500 milliseconds"; for the custom speech rate tag " <speed value="fast" / > The device extracts the factor parameter "speech speed sub-label, speed value is fast".

[0085] Step S205: Integrate the parameters of each factor to obtain the neutral abstract label.

[0086] It should be noted that after extracting all factor parameters, the device can integrate these parameters into a structured neutral abstract label object. This neutral abstract label object contains a label type field and the corresponding parameter field, and does not contain any vendor-specific syntax. The neutral abstract label is independent of any specific speech synthesis engine and can be losslessly transferred and converted between different manufacturers.

[0087] In practical implementation, the device can assemble various factor parameters according to a preset data structure. For example, emotion parameters can be encapsulated as an emotion field, speech rate parameters as a speech rate field, and pause parameters as a pause duration field. The integrated neutral abstract label serves as a standardized intermediate representation, which can be processed in subsequent rule mapping processes.

[0088] This embodiment can automatically select the optimal parsing by enumerating multiple candidate parsing results and combining a multi-dimensional scoring mechanism, effectively solving the parsing error problem caused by ambiguous dates and improving the accuracy of date normalization. At the same time, by using preset neutral label factors to perform fine-grained abstraction of custom SSML labels, heterogeneous custom labels are unified into structured neutral abstract labels, providing standardized input for subsequent cross-engine mapping.

[0089] Based on the first and second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the first and second embodiments described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the SSML tag processing method for speech synthesis in this application.

[0090] In this embodiment, to specifically illustrate how to apply configuration mapping rules to perform vendor adaptation mapping on the aforementioned neutral standard date entity and neutral abstract label, step S30 specifically includes: steps S301~S303: Step S301: Match the configuration mapping rules corresponding to the target speech synthesis engine in the rule configuration library. The rule configuration library pre-stores the configuration mapping rules between the engine specification language structure and the preset neutral standard date structure corresponding to different speech synthesis engines.

[0091] It should be noted that the rule configuration library can be a database or configuration file system that centrally stores mapping rules, pre-stores mapping rules for multiple speech synthesis engines. Each TTS engine from different vendors can correspond to an independent set of configuration mapping rules. These configuration mapping rules can be used to describe the correspondence between the proprietary SSML tag syntax specifications adopted by the corresponding TTS engine, the format specifications of the target text to be read, and the aforementioned neutral standard date structure.

[0092] In the implementation, the configuration mapping rules in the rule configuration library can be indexed and stored according to engine identifiers (such as engine name and engine version number). After the device obtains the currently selected target TTS engine, it uses the engine's identifier as the key to perform a matching query in the rule configuration library to obtain the configuration mapping rule uniquely corresponding to that engine. If there is no configuration mapping rule matching the target speech synthesis engine in the rule configuration library, the aforementioned degradation strategy is triggered, and standard-compatible SSML tags are directly generated.

[0093] Step S302: Determine the target date mapping rule and the target tag mapping rule according to the configuration mapping rule.

[0094] It should be noted that a configuration mapping rule can contain two independent sub-rules: the target date mapping rule and the target tag mapping rule.

[0095] The target date mapping rules are used to define the conversion method from the neutral standard date structure to the target read-aloud text, including the order of date fields (such as year-month-day, month-day-year, day-month-year), the type of date separator (such as Chinese characters, forward slash, hyphen), and the numerical representation of month and day (such as whether to include leading zeros, whether to use English month names), etc.

[0096] The target tag mapping rules are used to define the conversion method from neutral abstract tags to target TTS engine private SSML tags, including the mapping relationship of tag names, the mapping relationship of tag attributes, and the assembly format of tag syntax.

[0097] In practice, the device parses the target date mapping rule and the target tag mapping rule from the acquired configuration mapping rules and stores them as data structures for subsequent execution. If either the target date mapping rule or the target tag mapping rule is missing from the configuration mapping rules, the corresponding mapping step will be marked as unexecutable, triggering a degradation strategy.

[0098] Step S303: Map the neutral standard date entity to a target reading text based on the target date mapping rule, and map the neutral abstract label to a private SSML label corresponding to the target speech synthesis engine based on the target label mapping rule.

[0099] It should be noted that the neutral standard date entity can be a structured semantic object, including fields such as year, month, and day. The target date mapping rule can be used to define how to combine the above fields into a text string that conforms to the reading habit of the target engine.

[0100] For example, for the neutral standard date entity {year = 2021, month = 4, day = 5}, if the target date mapping rule is the Chinese format "{year} year {month} month {day} day", the mapping result is "April 5, 2021"; if the target date mapping rule is the American English format "{month} {day}, {year}", the mapping result is "April 5, 2026".

[0101] The neutral abstract label can also be a structured semantic object, including a label type and factor parameters. The target label mapping rule can be used to define how to convert the type and parameters in the neutral abstract label into a private SSML syntax that can be recognized by the target engine.

[0102] For example, for the neutral pause label {type = pause, duration = 500 milliseconds}, if the target label mapping rule is the syntax of manufacturer A " <break time="\"{duration}ms\" / ">The mapping result is " <break time="\"500ms\" / ">If the target label mapping rule is vendor B's syntax "{{pause:{duration}}}", then the mapping result is "{{pause:500}}".

[0103] In practice, the device can iterate through all neutral standard date entities and neutral abstract tags to be converted, and call the corresponding mapping rules to perform the conversion operation to obtain the converted target text fragment and private SSML tag fragment.

[0104] Furthermore, to illustrate specifically how to concatenate the target text to be read aloud and the private SSML tags obtained from the aforementioned conversion to obtain target-specific SSML tags adapted to the target TTS engine, step S40 also includes: steps S401~S403: Step S401: Determine the target splicing method according to the configuration mapping rules.

[0105] It should be noted that the target concatenation method is used to determine how to assemble the free text, the target read-aloud text fragment, and the private SSML tag fragment in their original order when generating the final target-specific SSML tags. Since different speech synthesis engines may have different requirements for tag nesting rules and the relative positions of tags and text, the target concatenation method is also stored as part of the configuration mapping rules.

[0106] In a practical implementation, the device can parse the target splicing method from the configuration mapping rules. This method can include the configuration of the connector between fragments (such as whether spaces or newlines need to be added), the configuration of the tag wrapping level (such as whether private SSML tags need to be wrapped in a specific parent tag), etc.

[0107] Step S402: Based on the target splicing method, splice the target text to be read aloud and the private SSML tag to obtain the splicing result.

[0108] It should be noted that the device can arrange and splice the free text fragment, the target text fragment obtained in step S33, and the private SSML tag fragment according to their order of appearance in the original text to be synthesized. The free text fragment retains its original content, the target text fragment replaces the original heterogeneous date entity position, and the private SSML tag fragment replaces the original custom SSML tag position.

[0109] For example, the original text is "The meeting time is set at..." <date> 2021-04-05< / date> , <pause / > "Please attend on time," after the aforementioned neutral mapping and vendor-specific mapping, the target text "April 5, 2026" and the private SSML tag "" are obtained. <break time="\"500ms\" / ">The concatenated result would be "The meeting is scheduled for April 5, 2026." <break time="\"500ms\" / ">Please be on time.

[0110] Step S403: Add semantic metadata annotations to the splicing result to generate target-specific SSML tags.

[0111] It should be noted that semantic metadata annotations can be invisible markers embedded in the target-specific SSML tags. They do not affect the normal reading of the speech synthesis engine and are only used to record the link information in the conversion process, facilitating subsequent debugging, tracking, and feedback correction. These semantic metadata annotations can include the conversion link representation from the heterogeneous date entity to the target read text and the conversion link representation from the custom SSML tag to the private SSML tag.

[0112] It should be understood that a transformation path representation records the complete transformation path from the original input to the final output. For date entities, a transformation path representation may include at least the text content of the original heterogeneous date entity, the structured content of the transformed neutral standard date entity, the identifier of the target date mapping rule used, and the final generated target text for reading aloud.

[0113] For custom SSML tags, the transformation chain representation can include at least the content of the original custom SSML tags, the content of the neutral abstract tags obtained after transformation, the target tag mapping rule identifier used, and the finally generated private SSML tags.

[0114] In practical implementation, the device can annotate semantic metadata in HTML comment format (such as "...").<!--date_trace:...--> ") or XML processing instruction format (such as "<?trace...?> Add the annotation "") to the appropriate position within the target-specific SSML tag, immediately adjacent to the corresponding transformation result segment. For example, for date mapping results, you can add comments before and after them: <!--date_trace: src="04-05-2021",mid={year=2021,month=4,day=5},rule="us_v1",dst="April 5,2026"--> .

[0115] It should be understood that the target-specific SSML tags with added semantic metadata annotations retain complete transformation chain information and can be parsed and executed normally by the target speech synthesis engine. This provides a crucial data foundation for subsequent audio quality feedback and adaptive rule correction.

[0116] Furthermore, this can be referenced here. Figure 4 This paper describes the entire process of the SSML tag processing method for speech synthesis in this application. Figure 4 This is a schematic diagram of the entire process of the SSML tag processing method for speech synthesis in this application.

[0117] Depend on Figure 4 As can be seen, firstly, the device acquires the text to be synthesized into speech, which contains heterogeneous date entities and custom SSML tags, and then extracts the heterogeneous date entities and custom SSML tags from it through the text parsing module (step S1).

[0118] Next, the device converts heterogeneous date entities into neutral standard date semantic structures and converts custom SSML tags into engine-independent neutral abstract tags (step S3).

[0119] Then, the device loads the corresponding configuration mapping rules according to the target speech synthesis engine and performs bidirectional mapping from neutral standard date entities to target text and from neutral abstract tags to private SSML tags (step S4).

[0120] After the mapping is completed, the device determines whether the mapping is successful (step S5). If the mapping is successful, the device renders and generates engine-specific SSML composite data and outputs it (step S6); if the mapping fails, a preset degradation strategy is triggered, and standard compatible SSML tags are generated and output (step S7, which is represented in the figure as "outputting standard compatible SSML after mapping failure").

[0121] Furthermore, you can also refer to this section. Figure 5 The device implementing the SSML tag processing method for speech synthesis described in this application is divided into a three-layer decoupled architecture. Figure 5 A device decoupling architecture block diagram for implementing the SSML tag processing method for speech synthesis in this application.

[0122] Depend on Figure 5 It can be seen that the device architecture can be divided into three layers: a heterogeneous access layer, a neutral semantic abstraction layer, and a configurable mapping rendering layer. The heterogeneous access layer is responsible for receiving inputs of multi-format text, custom SSML tags, and multilingual regional texts, completing the unified access and preliminary parsing of raw data.

[0123] The neutral semantic abstraction layer, as the core interpreter, performs standard semantic modeling on heterogeneous date entities, converts custom SSML tags into a set of neutral SSML abstract tags, and performs unified semantic storage to form a neutral semantic representation that is independent of the specific engine.

[0124] The configurable mapping rendering layer loads multi-dimensional rule configurations, executes engine-private tag mapping and text format rendering, and is equipped with degradation and fault tolerance processing. Finally, it outputs target engine-specific SSML composite data (output end: engine-specific SSML composite data / standard compatible SSML fallback data).

[0125] This embodiment enables dynamic mapping from neutral semantics to multi-engine private syntax through a rule configuration library. When adding a new engine, only the configuration needs to be extended without modifying the code logic. Furthermore, by adding semantic metadata annotations containing the complete transformation chain to the output results, the transformation process is made traceable and debuggable, laying the foundation for subsequent adaptive optimization of the system.

[0126] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the SSML tag processing method for speech synthesis in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0127] Furthermore, this application also provides an SSML tag processing device for speech synthesis, such as... Figure 6 As shown, Figure 6 This is a schematic diagram of the module structure of the SSML tag processing device for speech synthesis according to this application. Figure 6 It is known that the device includes: The text parsing module 601 is used to acquire the text to be synthesized into speech and to parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags. The normalization conversion module 602 is used to convert the heterogeneous date entity according to a preset neutral standard date structure and convert the custom SSML tag into a neutral abstract tag; The rule mapping module 603 is used to obtain the corresponding configuration mapping rules according to the target speech synthesis engine, and to map the converted neutral standard date entity to the target reading text based on the configuration mapping rules, and to map the neutral abstract tag to the private SSML tag corresponding to the target speech synthesis engine. The rendering output module 604 is used to concatenate the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag. The target-specific SSML tag is used to guide the target speech synthesis engine to perform speech synthesis.

[0128] This embodiment achieves decoupling of date text format and SSML tags from specific speech synthesis engines by converting heterogeneous date entities into neutral standard date structures, converting custom SSML tags into neutral abstract tags, and then performing reverse mapping based on the target speech synthesis engine's configuration mapping rules. The business system only needs to maintain one set of neutral semantic expressions, eliminating the need to write multiple sets of business code for different vendors. Simultaneously, a preset degradation strategy automatically generates standard-compatible SSML when mapping fails, ensuring the basic availability of the speech synthesis service.

[0129] This application also provides an SSML tag processing device for speech synthesis, the SSML tag processing device for speech synthesis includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the SSML tag processing method for speech synthesis in the first embodiment described above.

[0130] The following is for reference. Figure 7 , Figure 7 This is a schematic diagram of the SSML tag processing device for speech synthesis according to this application. The SSML tag processing device for speech synthesis in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, personal digital assistants (PDAs), tablet computers (PADs), portable media players (PMPs), etc., as well as fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The SSML tagging device for speech synthesis shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0131] like Figure 7 As shown, the SSML tag processing device for speech synthesis may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the SSML tag processing device for speech synthesis. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the speech synthesis-oriented SSML tag processing device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a speech synthesis-oriented SSML tag processing device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0132] The SSML tagging device for speech synthesis provided in this application, employing the SSML tagging method for speech synthesis described in the above embodiments, can solve the technical problems of SSML tagging for speech synthesis. Compared with the prior art, the beneficial effects of the SSML tagging device for speech synthesis provided in this application are the same as those of the SSML tagging method for speech synthesis provided in the above embodiments, and other technical features in this SSML tagging device for speech synthesis are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0133] This application also provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the SSML tag processing method for speech synthesis in the above embodiments.

[0134] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0135] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described SSML tag processing method for speech synthesis, thereby solving the technical problems of the SSML tag processing method for speech synthesis. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the SSML tag processing method for speech synthesis provided in the above embodiments, and will not be repeated here.

[0136] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other elements in the process, method, article, or system that includes that element.

[0137] The above embodiment numbers are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. They are only some embodiments of this application and do not limit the scope of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of this application.< / break> < / break> < / break> < / break>

Claims

1. A method for processing SSML tags for speech synthesis, characterized in that, The method includes: Obtain the text to be synthesized into speech, and parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags. The heterogeneous date entities are converted according to a preset neutral standard date structure, and the custom SSML tags are converted into neutral abstract tags; The corresponding configuration mapping rules are obtained according to the target speech synthesis engine, and the neutral standard date entity is mapped to the target reading text based on the configuration mapping rules, and the neutral abstract tag is mapped to the private SSML tag corresponding to the target speech synthesis engine. The target text to be read aloud is concatenated with the private SSML tag to generate a target-specific SSML tag, which is used to guide the target speech synthesis engine to perform speech synthesis.

2. The method as described in claim 1, characterized in that, The semantic parsing result also includes free text, and the step of converting the heterogeneous date entities according to a preset neutral standard date structure includes: The heterogeneous date entities are converted according to the preset neutral standard date structure to obtain several candidate neutral standard conversion entities; Based on the free text, each of the candidate neutral standard transformation entities is scored according to a preset scoring dimension, which includes at least two of the following: regional custom dimension, contextual semantic dimension, and timeliness dimension. Based on the scoring results, the candidate neutral standard conversion entity with the highest score is determined as the neutral standard date entity.

3. The method as described in claim 2, characterized in that, The step of scoring each candidate neutral standard transformation entity based on the free text according to a preset scoring dimension includes: Based on the regional configuration information of the target speech synthesis engine, determine the regional habit score of each of the candidate neutral standard conversion entities; Based on the free text, determine the context words around the heterogeneous date entity in the text to be synthesized, and determine the context semantic score of each of the candidate neutral standard transformation entities according to the context words; The timeliness score of each candidate neutral standard conversion entity is determined based on the time distance between the current date and the corresponding date of each candidate neutral standard conversion entity.

4. The method as described in claim 2, characterized in that, The step of converting the custom SSML tag into a neutral abstract tag includes: Extract the corresponding factor parameters from the custom SSML tags according to the preset neutral label factors. The preset label factors include: sentiment sub-tags, speech rate sub-tags, pause sub-tags, intonation sub-tags, and sentence break tags. By integrating the parameters of each factor, the neutral abstract label is obtained.

5. The method as described in claim 1, characterized in that, The steps of obtaining the corresponding configuration mapping rules based on the target speech synthesis engine, mapping the converted neutral standard date entities to the target read-aloud text based on the configuration mapping rules, and mapping the neutral abstract tags to the private SSML tags corresponding to the target speech synthesis engine include: In the rule configuration library, the configuration mapping rules corresponding to the target speech synthesis engine are matched. The rule configuration library contains configuration mapping rules between the engine specification language structure and the preset neutral standard date structure corresponding to different speech synthesis engines. The target date mapping rule and the target tag mapping rule are determined based on the configured mapping rules. Based on the target date mapping rule, the neutral standard date entity is mapped to the target read-aloud text, and based on the target tag mapping rule, the neutral abstract tag is mapped to the private SSML tag corresponding to the target speech synthesis engine.

6. The method as described in claim 5, characterized in that, The step of concatenating the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag includes: The target splicing method is determined according to the configured mapping rules; The target text to be read aloud and the private SSML tag are concatenated based on the target concatenation method to obtain the concatenation result; Add semantic metadata annotations to the concatenation result to generate target-specific SSML tags; The semantic metadata annotation includes the conversion link representation from the heterogeneous date entity to the target read-aloud text and the conversion link representation from the custom SSML tag to the private SSML tag.

7. The method as described in claim 1, characterized in that, Following the step of obtaining the corresponding configuration mapping rules based on the target speech synthesis engine, the method further includes: Determine if there are any instances of target text conversion failure or private SSML tag conversion failure; If so, a preset downgrade strategy is executed to generate standard compatible SSML tags, which are used to guide the target speech synthesis engine to perform speech synthesis. The preset degradation strategy includes: outputting the neutral standard date entity as the original text format or outputting the remaining text after removing the neutral abstract tags that cannot be mapped.

8. An SSML tagging processing device for speech synthesis, characterized in that, The device includes: The text parsing module is used to acquire the text to be synthesized into speech and to parse the text to be synthesized into speech to obtain semantic parsing results. The semantic parsing results include at least: heterogeneous date entities and custom SSML tags. The normalization conversion module is used to convert the heterogeneous date entities according to a preset neutral standard date structure and convert the custom SSML tags into neutral abstract tags; The rule mapping module is used to obtain the corresponding configuration mapping rules according to the target speech synthesis engine, and to map the converted neutral standard date entity to the target reading text based on the configuration mapping rules, and to map the neutral abstract tag to the private SSML tag corresponding to the target speech synthesis engine. The rendering output module is used to concatenate the target text to be read aloud with the private SSML tag to generate a target-specific SSML tag. The target-specific SSML tag is used to guide the target speech synthesis engine to perform speech synthesis.

9. An SSML tag processing device for speech synthesis, characterized in that, The device includes: a memory, a processor, and an SSML tagging program for speech synthesis stored in the memory and executable on the processor, wherein the SSML tagging program for speech synthesis, when executed by the processor, implements the SSML tagging method for speech synthesis as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores an SSML tag processing program for speech synthesis, which, when executed by a processor, implements the SSML tag processing method for speech synthesis as described in any one of claims 1 to 7.

Images