A podcast automatic generation method and related device

Abstract

The application provides a podcast automatic generation method and related device, and relates to the technical field of software. The method comprises the following steps: obtaining a target text to be processed through webpage analysis; obtaining a prompt word used for indicating a dialogue logic, and inputting the prompt word and the target text into a large language model, so that the large language model converts the target text into a script text based on the prompt word; and performing voice synthesis on the script text based on a timbre cloning to obtain a podcast audio. The application can realize end-to-end automatic generation from a non-structured webpage text to a structured multi-role dialogue audio, enhance the interaction and scene of the content, and meet the consumption demand of users for diversified and personalized audio content.

Description

Technical Field

[0001] This application relates to the field of software technology, and in particular to a method and apparatus for automatically generating podcasts. Background Technology

[0002] With the popularization of mobile internet and the widespread use of smart terminal devices, users are increasingly diversifying their ways of obtaining information, among which consuming news and information through podcasts has become an important trend.

[0003] Currently, podcast audio mainly relies on traditional text-to-speech (TTS) technology. The system usually reads the parsed web page text linearly and one-way word by word. Its processing focuses on converting character sequences into acoustic signals while keeping the original narrative structure unchanged. Therefore, the generated podcast audio often presents a monotonous linear reading, lacking interactivity and a sense of context. Summary of the Invention

[0004] In view of the above problems, this application provides a method and related apparatus for automatically generating podcasts, so as to achieve the purpose of generating podcast audio according to users' personalized needs. The specific solution is as follows:

[0005] The first aspect of this application provides a method for automatically generating podcasts, the method comprising:

[0006] Obtain the target text to be processed by parsing the webpage;

[0007] Obtain cue words used to indicate dialogue logic, and input the cue words and the target text into a large language model, so that the large language model converts the target text into script text based on the cue words;

[0008] The podcast audio is obtained by synthesizing the speech of the script text based on timbre cloning.

[0009] In one possible implementation, obtaining the target text to be processed through webpage parsing includes:

[0010] Receive a webpage link and parse it to obtain the original text under the webpage link;

[0011] The target text is obtained by removing interfering content from the original text and then performing regularization.

[0012] In one possible implementation, obtaining the prompt words used to indicate the dialogue logic includes:

[0013] Obtain the script configuration, which includes dialogue characters and the script's emotional style;

[0014] The cue words are assembled based on the dialogue characters and the emotional style of the script.

[0015] In one possible implementation, the script text includes the dialogue text and emotional style of the dialogue characters in a single turn of conversation, and the step of synthesizing the podcast audio from the script text based on voice cloning includes:

[0016] Obtain the voiceprint features of the dialogue characters;

[0017] Based on the voiceprint features and the emotional style of the dialogue, the dialogue text is synthesized into an audio segment;

[0018] The audio segments are spliced ​​together to form the podcast audio based on the dialogue sequence of the single-turn conversation.

[0019] In one possible implementation, the podcast automatic generation method further includes:

[0020] Obtain background audio that matches the emotional style of the script, and mix the background audio and the podcast audio.

[0021] In one possible implementation, the mixing process for the background audio and the podcast audio includes:

[0022] Determine the vocal envelope parameters of the podcast audio;

[0023] The gain coefficient of the background audio is adjusted according to the human voice envelope parameters, and the gain coefficient is used to adjust the playback volume of the background audio.

[0024] A second aspect of this application provides an automatic podcast generation device, the automatic podcast generation device comprising:

[0025] The text parsing module is used to obtain the target text to be processed by parsing web pages;

[0026] The podcast generation module is used to obtain cue words that indicate the dialogue logic, and input the cue words and the target text into a large language model, so that the large language model converts the target text into script text based on the cue words; and performs speech synthesis on the script text based on timbre cloning to obtain podcast audio.

[0027] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the podcast automatic generation method described in the first aspect or any implementation thereof.

[0028] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:

[0029] The memory is used to store computer programs;

[0030] The processor is used to execute the computer program to enable the electronic device to implement the podcast automatic generation method of the first aspect or any implementation thereof.

[0031] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the podcast automatic generation method described in the first aspect or any implementation thereof.

[0032] By employing the aforementioned technical solutions, this application provides a method and related apparatus for automatically generating podcasts. The method includes: obtaining target text to be processed through webpage parsing; obtaining prompt words to indicate dialogue logic, and inputting the prompt words and target text into a large language model, so that the large language model converts the target text into script text based on the prompt words; and synthesizing podcast audio from the script text using voice cloning. In this application, by using prompt words to indicate dialogue logic, the large language model can understand and execute specific script logic, converting linear text into script text with interactivity and a sense of scene. Furthermore, voice cloning can synthesize the script text into podcast audio with specific voiceprint characteristics, thereby achieving end-to-end automated generation from unstructured webpage text to structured, multi-role dialogue audio, enhancing the interactivity and scene-based nature of the content, and meeting users' consumption needs for diverse and personalized audio content. Attached Figure Description

[0033] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0034] Figure 1 A flowchart illustrating an automatic podcast generation method provided in this application embodiment;

[0035] Figure 2 This is a partial flowchart illustrating a podcast automatic generation method provided in an embodiment of this application;

[0036] Figure 3 This is another schematic flowchart of a podcast automatic generation method provided in an embodiment of this application;

[0037] Figure 4 This is another schematic flowchart of a podcast automatic generation method provided in an embodiment of this application;

[0038] Figure 5 This is another schematic flowchart of a podcast automatic generation method provided in an embodiment of this application;

[0039] Figure 6 This is a schematic diagram of the structure of an automatic podcast generation device provided in an embodiment of this application;

[0040] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0041] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0042] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0043] The terms "first," "second," etc., used in the specification, claims, 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 terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0044] To address the problem that traditional podcast audio is presented as monotonous linear reading, lacking interactivity and a sense of context, this application provides a method for automatically generating podcasts. The following is a detailed description of this method with reference to the accompanying drawings.

[0045] See Figure 1 , Figure 1 This is a flowchart illustrating an automatic podcast generation method provided in an embodiment of this application. Figure 1 As shown in the figure, the podcast automatic generation method provided in this application embodiment may include steps S101 to S103, which are described in detail below.

[0046] S101, obtain the target text to be processed through webpage parsing.

[0047] In this embodiment of the application, for a webpage link entered by the user, such as a news URL (Uniform Resource Locator), an automated script or parsing library can be used to initiate a network request to obtain the text content under the webpage link, which is then used as the target text to be processed.

[0048] In one possible implementation, to achieve efficient conversion from unstructured web page data to standardized text data, embodiments of this application utilize regularization rules to perform secondary processing on the original text. This eliminates the interference of web page-specific formats on semantic understanding, thereby ensuring that the text input into the large language model has extremely high purity and coherence, significantly improving the listenability of podcast audio. See also Figure 2 , Figure 2 This is a partial flowchart illustrating a podcast automatic generation method provided in an embodiment of this application. Figure 2 As shown in the embodiment of this application, an automatic podcast generation method is provided, wherein step S101, "obtaining the target text to be processed through webpage parsing", may include steps S201 to S202, which are described in detail below.

[0049] S201: Receive the webpage link and parse it to obtain the original text under the webpage link.

[0050] In this embodiment, a webpage link input by the user can be obtained through a user interface. This webpage link points to an internet page containing news or article content to be processed. A webpage parsing tool is invoked to initiate an HTTP request to the target URL corresponding to the webpage link, obtain the Hypertext Markup Language (HTML) source code of the page, and extract the text content based on the Document Object Model (DOM) tree structure, which is then used as the raw text.

[0051] Specifically, the Python third-party library trafilatura is used as the parsing engine. This engine can automatically identify the structure of most news websites that do not require login. Additionally, a fallback strategy for HTML DOM tree analysis is configured to prevent abnormal website structures. During parsing, each node of the DOM tree is traversed, calculating the text density and tag weight of each node to identify the area with the highest text density as the core text. All text content within this area is then concatenated to form the original text. For example, when a user enters a link to a tech news website, high-noise areas such as the top navigation bar, side banner ads, and bottom copyright notices are automatically ignored, and only the main news content in the middle is extracted as the original text. This combination of statistical text density assessment and structured HTML analysis allows for precise location of effective information blocks, avoiding reliance on manual filtering.

[0052] S202: Remove interfering content from the original text and obtain the target text through regularization.

[0053] In this embodiment, the obtained original text can be further processed to remove residual non-text elements, such as residual text in densely linked areas, script code snippets, or irrelevant metadata and other interfering content. Then, for the original text after removing interfering content, a format standardization operation is performed using predefined regular expression rules. This specifically includes removing redundant line breaks, consecutive spaces, special symbols (such as unreadable characters or garbled text), and punctuation combinations that do not conform to natural language habits. This ensures that the text fed into the large language model is clean, coherent, and conforms to natural language processing standards. For example, if the original text contains a large number of blank lines caused by webpage layout or heading tags represented by ###, regularization processing will replace these formatting symbols with standard paragraph separators or delete them directly, while merging abnormal spaces between scattered numbers and units.

[0054] In this embodiment of the application, the above-mentioned operations such as removing interference content and regularization can significantly improve the signal-to-noise ratio of the input data, effectively avoiding the risk of misunderstanding or deterioration in the generation quality of the large language model due to format confusion or noise interference, and ensuring the generation of high-quality dialogue scripts in the future.

[0055] S102, obtain the cue words used to indicate the dialogue logic, and input the cue words and target text into the large language model so that the large language model can convert the target text into script text based on the cue words.

[0056] In this embodiment, the prompt words are structured text containing a set of instructions with specific dialogue logic, used to guide the large language model to generate content according to corresponding logical constraints. These prompt words may include dialogue role settings, script emotional style, structural design requirements, and language style specifications. Specifically, dialogue role settings define the identities (e.g., host, guest), personality traits, and relationships of the characters participating in the dialogue; script emotional style specifies the desired tone and atmosphere (e.g., humorous, serious, suspenseful); structural design requirements define the logical flow of the script's opening remarks, key topic discussions, and conclusion; and language style specifications constrain the colloquialism of the output text, the use of filler words, and punctuation format.

[0057] In practical applications, prompts can be dynamically assembled based on the user's configuration parameters. For example, if the user selects a two-person dialogue and a humorous style, the corresponding role templates (such as a female host responsible for controlling the conversation and asking questions, and a male guest responsible for explaining professional knowledge) and style instructions (such as using more colloquial expressions, allowing repetition, and adding filler words) will be embedded into the prompts.

[0058] Furthermore, the large language model is a deep neural network model based on the Transformer architecture, possessing powerful contextual understanding and natural language generation capabilities. In this embodiment, after obtaining the prompt words, the prompt words and target text are input into the large language model. Based on the constraints in the prompt words, the large language model semantically reconstructs the linear target text, transforming it into a script text with interactivity, tone annotation, and scene awareness. This script text is a structured multi-turn dialogue script. In this process, the large language model not only extracts key information from the target text but also utilizes its internal knowledge base to supplement background knowledge and construct metaphors or examples to ensure appropriate information density and ease of understanding. For example, the large language model can rewrite a dry description of technical parameters into a vivid discussion between two characters about how the technology affects daily life, and automatically add emotional tags (such as excitement and doubt).

[0059] In this embodiment of the application, by introducing a large language model as a scriptwriting engine, the automatic conversion from unstructured web page text to structured script text is realized.

[0060] S103 uses voice cloning to synthesize podcast audio from script text.

[0061] In this embodiment, voice cloning utilizes deep learning technology to extract speaker voiceprint features from reference audio and applies them to the speech synthesis process to generate speech signals with specific voiceprint features. After obtaining the script text, the script text is parsed to identify the dialogue role to which each line belongs. For each dialogue role, its corresponding voiceprint features are obtained. For example, if the user selects a custom voiceprint, the user can upload a short audio clip of human voice reference, and the voiceprint extractor extracts the voiceprint features from it; if the user selects a preset voiceprint, the voiceprint features of the corresponding role are retrieved from the database. It should be noted that voiceprint features are multi-dimensional vector data that can uniquely represent the voiceprint attributes of a specific speaker, and are usually referred to as voiceprint embedding vectors.

[0062] Furthermore, an end-to-end speech synthesis model, such as an architecture based on a diffusion model or an autoregressive model, is used to synthesize the dialogue content of the script text into time-domain waveform audio segments. In this process, parallel processing of multiple characters is supported, that is, the script text is traversed, and when different characters' lines are encountered, the voiceprint feature vectors used are dynamically switched, thereby enabling a single model to support output from multiple characters, and each character's voice has unique recognizability and realism.

[0063] Finally, following the dialogue sequence specified in the script, the generated audio segments are spliced ​​together, and natural silence intervals are inserted at dialogue transitions to simulate human pauses in thought, thus obtaining podcast audio. This podcast audio is an audio file containing dialogue between multiple characters and corresponding emotional rhythms.

[0064] In this embodiment of the application, by combining timbre cloning technology with structured scripts, not only can the automated production of podcast audio be achieved, but also the customization of timbre and style breaks through the limitations of traditional speech synthesis, which has a single timbre and lacks emotion, and significantly improves the user's immersive listening experience.

[0065] In one possible implementation, to enhance the listenability and user immersion of automatically generated content, this embodiment can assemble prompt words based on script configuration. The dialogue characters provide the main structure of the generated text for the large language model, while the script's emotional style determines the rhythm and emotional direction of the language. The prompt words assembled based on these two elements not only constrain the format standardization of the large language model's output but also stimulate the model's ability to switch between specific styles, thereby enabling the mass production of script texts with diverse styles and distinct characters without human intervention. See also Figure 3 , Figure 3 This is another schematic flowchart illustrating an automatic podcast generation method provided in an embodiment of this application. Figure 3 As shown in the embodiment of this application, an automatic podcast generation method is provided, wherein step S102, "obtaining prompt words for indicating dialogue logic", may include steps S301 to S302, which are described in detail below.

[0066] S301, Get script configuration, which includes dialogue characters and script emotional style.

[0067] In this embodiment, the script configuration is a set of parameters pre-configured by the user before initiating the podcast generation task. It includes dialogue roles and script emotional style. The dialogue roles can specifically include the number of people participating in the podcast dialogue, their identities, and the interaction relationships between them. For example, it can be set as a two-person dialogue mode, including a host role responsible for guiding the topic and a guest role responsible for providing professional knowledge; or it can be set as a multi-person roundtable discussion mode, including three or more roles with different viewpoints. The script emotional style can include the overall linguistic tone and emotional color of the podcast audio presentation, including styles such as humorous, serious and professional, suspenseful and thrilling, or lighthearted and casual.

[0068] In practical applications, script configuration can be obtained by receiving user selection instructions through a graphical user interface, or it can be read from a historical user preference database. This application embodiment does not limit this.

[0069] S302, assemble cue words based on dialogue characters and the emotional style of the script.

[0070] In this embodiment, the assembled prompt typically includes multiple units such as a task overview, original content placeholders, content and structure requirements, language style specifications, and dialogue design logic. In the task overview unit, the large language model is explicitly instructed to generate a podcast script based on the provided target text. In the content and structure requirements unit, the structural framework of the opening remarks, key topic discussions, and closing remarks is defined according to the dialogue roles, and the information density and knowledge introduction methods are specified. In the language style specifications unit, corresponding interjections, filler words (such as "indeed," "uh," "actually"), and sentence structure features are injected according to the script's emotional style. For example, if the style is humorous, the prompt will include instructions allowing moderate teasing, using colloquial expressions, and creating contrast; if the style is serious, the instructions emphasize logical rigor and clear terminology explanations.

[0071] By dynamically assembling cue words based on dialogue characters and the emotional style of the script, the large language model is given a clear screenwriter identity and code of conduct, and can reconstruct linear unstructured text into script text with multi-turn interaction, rich emotions and logical coherence.

[0072] In one possible implementation, the script text includes the dialogue text and emotional style of the characters in a single turn of conversation. The dialogue text, emotional style, and voiceprint features of the respective characters are used as input to a speech synthesis model, which automatically generates corresponding audio segments, which are then spliced ​​together to obtain podcast audio. See also Figure 4 , Figure 4 This is another schematic flowchart illustrating an automatic podcast generation method provided in an embodiment of this application. Figure 4 As shown in the embodiment of this application, an automatic podcast generation method is provided, wherein step S103, "deriving podcast audio by synthesizing speech from script text based on timbre cloning", may include steps S401 to S403, which are described in detail below.

[0073] S401, Obtain the voiceprint characteristics of the dialogue character.

[0074] In this embodiment, reference audio containing the voices of dialogue characters is received, either uploaded by a user or retrieved from a preset library. The duration of this reference audio is typically between 3 and 10 seconds. Then, a pre-trained voiceprint coding model is used to extract features from this reference audio, mapping the temporal audio signal into a fixed-length vector in a high-dimensional space, thereby obtaining the corresponding voiceprint features. These voiceprint features serve as conditional inputs to the speech synthesis model, used to lock the timbre base of the generated speech, ensuring that the synthesized speech has the same timbre recognizability as the reference audio. For example, when a user wants a male guest in a podcast to use the voice of a well-known host, an audio clip of that host is obtained, processed, and the corresponding voiceprint features are generated. All subsequent dialogue text labeled as belonging to the male guest will be synthesized based on these voiceprint features.

[0075] By using this method of dynamically extracting voiceprint features based on reference audio, zero-sample or few-sample cloning of timbre is achieved, enabling each dialogue character in podcast audio to have an independent and customizable personalized timbre.

[0076] S402 synthesizes dialogue text into audio segments based on voiceprint features and dialogue emotional style.

[0077] In this embodiment, the obtained voiceprint features, dialogue emotional style, and dialogue text are input into a speech synthesis model, which then converts the dialogue text into audio segments with the voiceprint features and dialogue emotional style. The voiceprint features control the timbre attributes of the audio segments, while the dialogue emotional style adjusts the rhythm, intonation, and pauses of the audio segments. The two work together to achieve refined control over speech expression.

[0078] For example, for dialogue text tagged as "excited," the fundamental frequency variation rate is increased and syllable duration is shortened during synthesis. Simultaneously, specific voiceprint features are combined to generate audio clips that are both consistent with the character's timbre and full of emotional tension. Conversely, for calm text, audio clips with a steady tone and moderate pace are generated. This joint modeling approach not only solves the problem of limited emotional expression in traditional TTS technology but also ensures the consistency and stability of timbre during multi-character switching, resulting in highly natural and engaging audio clips.

[0079] S403 splices audio segments into podcast audio using the dialogue sequence of a single turn.

[0080] In this embodiment, the dialogue sequence is used to characterize the chronological order of each dialogue turn in the script text. To integrate discrete single-turn dialogue audio segments into a continuous and smooth podcast audio, the audio segments of each dialogue character can be read and arranged sequentially according to the dialogue sequence. To simulate the natural rhythm of real human dialogue and avoid the abruptness caused by mechanical splicing, a preset silence interval can be automatically inserted between two adjacent audio segments. The duration of this silence interval can be dynamically adjusted according to the dialogue context, typically set between 300 and 500 milliseconds to simulate the pauses of human thinking or breathing. For example, in a heated debate scene, the silence interval between audio segments can be shortened to 200 milliseconds to reflect a sense of urgency; while in a relaxed chat scene, it can be extended to 600 milliseconds to create a soothing atmosphere.

[0081] Finally, all audio segments, after being arranged chronologically and filled in with gaps, are seamlessly spliced ​​together to generate a complete podcast audio. This process effectively eliminates the auditory fragmentation caused by switching between multiple voices and text breaks, reconstructs realistic dialogue scenes, and significantly improves the user's listening experience.

[0082] In one possible implementation, to ensure clear vocals in podcast audio while preserving the atmosphere, this application embodiment uses a ducking technique for mixing. To this end, this application embodiment provides an automatic podcast generation method, which further includes the following steps:

[0083] Obtain background audio that matches the emotional style of the script, and mix the background audio and podcast audio.

[0084] In this embodiment, the background audio is non-human vocal music or sound effects that enhance the podcast atmosphere and emotional impact. It can be obtained from data sources such as a locally pre-built music library, a cloud audio resource pool, or real-time generated synthesized music. Specifically, the script's emotional style is used as a search keyword or feature vector to match and filter data from the data source to determine the matching background audio. Specifically, a mapping table between different script emotional styles and background audio types can be pre-established. When the script's emotional style is identified as suspenseful, low-frequency, slow-paced background audio with a sense of tension is automatically retrieved; when the script's emotional style is identified as cheerful, upbeat, high-pitched background audio is matched. Through this semantic style-based automatic matching mechanism, the emotional tone of the background audio is ensured to be highly consistent with the logical direction of the dialogue content, thereby constructing an immersive sense of scene at the auditory level.

[0085] Furthermore, based on ducking technology, background audio and podcast audio are mixed, and the volume of background audio is automatically reduced when human voices are detected.

[0086] In one possible implementation, to ensure a perfect balance between the prominence of the human voice and the background atmosphere in different content scenarios, and to enhance the listening comfort and professionalism of podcast audio, this embodiment dynamically adjusts the gain coefficient of the background audio by calculating the human voice envelope parameters. See also Figure 5 , Figure 5 This is another schematic flowchart illustrating an automatic podcast generation method provided in an embodiment of this application. Figure 5 As shown in the embodiment of this application, an automatic podcast generation method is provided, wherein the above-mentioned step "mixing background audio and podcast audio" may include steps S501 to S502, which are described in detail below.

[0087] S501, determine the vocal envelope parameters of the podcast audio.

[0088] In this embodiment, the human voice envelope parameter is used to characterize the trend of the amplitude of the human voice signal in podcast audio changing over time. It can be obtained by recursively calculating the instantaneous amplitude of the human voice audio signal using a smoothing algorithm. Specifically, the human voice envelope parameter can be calculated according to the following formula (1):

[0089] (1);

[0090] in, express Voice envelope parameters at any given time; This represents the preset smoothing coefficient, which typically ranges from 0 to 1. It is used to control the response speed of the envelope to signal changes and avoid sudden volume changes that could cause auditory discomfort. express The amplitude of the human voice audio signal at a given moment; express The human voice envelope parameters at any given time.

[0091] By introducing the synergistic effect of a smoothing coefficient and historical states, the process of determining the human voice envelope parameters avoids the jagged abrupt changes caused by directly taking absolute values, ensuring that the extracted energy curve can truly reflect the overall loudness profile of the human voice, and providing a stable input for the subsequent calculation of the gain coefficient.

[0092] S502 adjusts the gain coefficient of the background audio based on the human voice envelope parameters. The gain coefficient is used to adjust the playback volume of the background audio.

[0093] In this embodiment, the gain coefficient is an amplitude scaling factor applied to the background audio, thereby dynamically lowering or raising the volume of the background music according to the current activity level of the human voice, achieving a dodging effect. It can be calculated based on the human voice envelope parameters combined with a preset dodging depth coefficient, noise threshold, and maximum preset volume through a nonlinear mapping relationship. Specifically, the gain coefficient can be calculated according to the following formula (2):

[0094] (2);

[0095] in, express Gain coefficient at time t; This indicates the maximum preset volume of the background audio before mixing, usually normalized to 1. This indicates the preset ducking depth coefficient, which determines the degree to which the background audio is suppressed when human voices appear; This indicates the preset noise threshold value, which is used to set the energy threshold for triggering the avoidance mechanism, preventing ambient noise or faint breathing sounds from mistakenly triggering a volume reduction.

[0096] For example, when the broadcaster starts speaking, the voice envelope parameter rises rapidly and exceeds the noise threshold. The growth rate of the denominator in the above formula (2) is slower than that of the numerator, causing the fraction value to approach 1, which in turn causes the gain coefficient to drop significantly and the volume of the background audio to decrease automatically. When the broadcaster pauses and the voice envelope parameter falls back below the noise threshold, the gain coefficient gradually rises back to the maximum volume of the background audio.

[0097] This application embodiment achieves fine-grained adaptive control of background audio volume. On the one hand, the existence of a noise threshold shields invalid signal interference, ensuring the integrity of background audio during silent periods; on the other hand, the dodging depth coefficient and the smoothed human voice envelope parameters work together to ensure the smoothness and naturalness of volume transitions, completely eliminating the abruptness of auditory transitions caused by traditional hard switching methods.

[0098] Based on the above description, the podcast automatic generation method provided in this application embodiment can realize end-to-end automatic generation from unstructured web page text to structured, multi-style, multi-role dialogue audio, enhance the interactivity and context of the content, and meet users' consumption needs for diverse and personalized audio content.

[0099] The above describes an automatic podcast generation method provided by the embodiments of this application. The following describes the apparatus for performing the above-described automatic podcast generation method.

[0100] See Figure 6 , Figure 6 This is a schematic diagram of a podcast automatic generation device provided in an embodiment of this application. Figure 6As shown in the figure, an automatic podcast generation device provided in this application includes:

[0101] Text parsing module 601 is used to obtain the target text to be processed through webpage parsing;

[0102] The podcast generation module 602 is used to obtain cue words that indicate the dialogue logic, and input the cue words and target text into the large language model so that the large language model can convert the target text into script text based on the cue words; and synthesize the podcast audio by performing speech synthesis on the script text based on timbre cloning.

[0103] In one possible implementation, the text parsing module 601, used to obtain the target text to be processed through webpage parsing, is specifically used for:

[0104] Receive web page links and parse them to obtain the original text under the web page links; remove the interference content in the original text and perform regularization to obtain the target text.

[0105] In one possible implementation, the podcast generation module 602, used to obtain cue words for indicating dialogue logic, is specifically used for:

[0106] Obtain the script configuration, which includes dialogue characters and the script's emotional style; assemble cue words based on the dialogue characters and the script's emotional style.

[0107] In one possible implementation, the script text includes the dialogue text and emotional style of the dialogue characters in a single turn, and a podcast generation module 602 is used to synthesize the podcast audio from the script text based on timbre cloning, specifically for:

[0108] Obtain the voiceprint features of the dialogue characters; synthesize the dialogue text into audio segments based on the voiceprint features and the emotional style of the dialogue; and splice the audio segments into podcast audio according to the dialogue sequence of a single turn.

[0109] In one possible implementation, podcast generation module 602 is also used for:

[0110] Obtain background audio that matches the emotional style of the script, and mix the background audio and podcast audio.

[0111] In one possible implementation, the podcast generation module 602, used for mixing background audio and podcast audio, is specifically used for:

[0112] Determine the vocal envelope parameters of the podcast audio; adjust the gain coefficient of the background audio based on the vocal envelope parameters. The gain coefficient is used to adjust the playback volume of the background audio.

[0113] It should be noted that the detailed functions of each module in the embodiments of this application can be found in the corresponding disclosure of the above-mentioned podcast automatic generation method embodiments, and will not be repeated here.

[0114] This application also provides an electronic device in its embodiments. See also... Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device in this embodiment may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 7 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0115] like Figure 7 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. When the electronic device is powered on, the RAM 703 also stores various programs and data required for the operation of the electronic device. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0116] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, memory cards, hard drives, etc.; and communication devices 709. Communication device 709 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0117] This application also provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the podcast automatic generation methods provided in this application.

[0118] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the podcast automatic generation methods provided in this application.

[0119] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0120] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0121] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0122] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for automatically generating podcasts, characterized in that, The podcast automatic generation method includes: Obtain the target text to be processed by parsing the webpage; Obtain cue words used to indicate dialogue logic, and input the cue words and the target text into a large language model, so that the large language model converts the target text into script text based on the cue words; The podcast audio is obtained by synthesizing the speech of the script text based on timbre cloning. The step of obtaining prompt words for indicating dialogue logic includes: obtaining script configuration, which includes dialogue characters and script emotional style; and assembling the prompt words based on the dialogue characters and the script emotional style. The podcast automatic generation method further includes: obtaining background audio that matches the emotional style of the script, and mixing the background audio and the podcast audio; The mixing process for the background audio and the podcast audio includes: determining the vocal envelope parameters of the podcast audio; adjusting the gain coefficient of the background audio based on the vocal envelope parameters, wherein the gain coefficient is used to adjust the playback volume of the background audio; wherein... , express Voice envelope parameters at any given time; Indicates the preset smoothing coefficient; express The amplitude of the human voice audio signal at a given moment; express Voice envelope parameters at any given time; , express Gain coefficient at time t; This indicates the maximum preset volume of the background audio before mixing; This indicates the preset dodge depth coefficient; This indicates the preset noise threshold value.

2. The podcast automatic generation method according to claim 1, characterized in that, The process of obtaining the target text to be processed through webpage parsing includes: Receive a webpage link and parse it to obtain the original text under the webpage link; The target text is obtained by removing interfering content from the original text and then performing regularization.

3. The podcast automatic generation method according to claim 1, characterized in that, The script text includes the dialogue text and emotional style of the dialogue characters in a single turn of conversation. The step of synthesizing the podcast audio from the script text based on voice cloning includes: Obtain the voiceprint features of the dialogue characters; Based on the voiceprint features and the emotional style of the dialogue, the dialogue text is synthesized into an audio segment; The audio segments are spliced ​​together to form the podcast audio based on the dialogue sequence of the single-turn conversation.

4. A podcast automatic generation device, characterized in that, The podcast automatic generation device includes: The text parsing module is used to obtain the target text to be processed by parsing web pages; The podcast generation module is used to acquire cue words that indicate dialogue logic, and input the cue words and the target text into a large language model, so that the large language model converts the target text into script text based on the cue words; and performs speech synthesis on the script text based on timbre cloning to obtain podcast audio; The podcast generation module for obtaining cue words to indicate dialogue logic is specifically used for: obtaining script configuration, which includes dialogue characters and script emotional style; and assembling the cue words based on the dialogue characters and the script emotional style. The podcast generation module is further configured to: acquire background audio that matches the emotional style of the script, and mix the background audio and the podcast audio. The podcast generation module, used for mixing the background audio and the podcast audio, is specifically configured to: determine the vocal envelope parameters of the podcast audio; adjust the gain coefficient of the background audio based on the vocal envelope parameters, wherein the gain coefficient is used to adjust the playback volume of the background audio; wherein... , express Voice envelope parameters at any given time; Indicates the preset smoothing coefficient; express The amplitude of the human voice audio signal at a given moment; express Voice envelope parameters at any given time; , express Gain coefficient at time t; This indicates the maximum preset volume of the background audio before mixing; This indicates the preset dodge depth coefficient; This indicates the preset noise threshold value.

5. A computer program product, characterized in that, Includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the podcast automatic generation method as described in any one of claims 1 to 3.

6. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program to enable the electronic device to implement the podcast automatic generation method as described in any one of claims 1 to 3.

7. A computer storage medium, characterized in that, The storage medium carries one or more computer programs that, when executed by an electronic device, enable the electronic device to implement the podcast automatic generation method as described in any one of claims 1 to 3.

Images