A lyric generation method, electronic device, and storage medium

By acquiring user music feature data and utilizing a multi-stage creation process driven by AI, personalized lyrics are generated, solving the problem of homogenized lyrics in existing technologies and improving user experience and the matching degree between lyrics and music.

CN123115458APending Publication Date: 2026-11-17HANGZHOU NETEASE CLOUD MUSIC TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611208008.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, lyric generation methods lack in-depth analysis of user music feature data, resulting in highly homogenized lyrics that fail to meet users' personalized needs and lead to a poor user experience.

Method used

By acquiring users' musical characteristic data and utilizing a multi-stage creation process driven by AI agents, lyrics with highly personalized imagery density, strict rhythmic control, and clear emotional arcs are generated. Combined with embedded metadata tags, this enables precise paragraph-level control over the downstream AI music generation engine.

Benefits of technology

It enables personalized lyric creation, enhances user experience, ensures high matching between lyric content and musical characteristics, generates strong emotional resonance, and avoids homogenization of skeletal structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115458A_ABST
    Figure CN123115458A_ABST
Patent Text Reader

Abstract

This disclosure presents a lyric generation method, electronic device, and storage medium. The method involves acquiring a user's music feature data; processing the music feature data to determine a target skeleton template; the target skeleton template indicating the paragraph structure of the lyrics; and based on the target skeleton template, combined with the music feature data, performing imagery anchoring processing to generate lyrics that carry images matching the music feature data, thereby meeting the user's personalized creation needs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of audio technology, specifically to a lyric generation method, an electronic device, and a storage medium. Background Technology

[0002] Currently, some applications offer lyrics generation functionality to support user music creation. However, these technologies typically utilize fixed, generic paragraph structure templates to generate lyrics, lacking in-depth analysis mechanisms for user musical characteristic data. This prevents the determination of suitable paragraph structures based on user characteristic data and also lacks personalized processing strategies that link user characteristic data with lyric content generation. Furthermore, because existing methods fail to effectively utilize acquired user characteristic data for personalized structure and content adaptation, the lyrics generated by terminal devices are highly homogenized in terms of paragraph structure and content expression, resulting in a waste of data processing resources, failure to meet users' personalized needs, and a poor user experience. Summary of the Invention

[0003] This disclosure provides a lyrics generation method, an electronic device, and a storage medium to generate lyrics that meet users' personalized needs and improve user experience.

[0004] In a first aspect, embodiments of this disclosure provide a lyrics generation method, including: Obtain user's music feature data; The music feature data is processed to determine the target skeleton template; the target skeleton template indicates the paragraph structure of the lyrics; Based on the target skeleton template, and combined with the music feature data, imagery anchoring processing is performed to generate lyrics that carry images that match the music feature data.

[0005] In one possible design approach, after generating lyrics that carry visuals matching the musical feature data, the method further includes: Add structure tags and production instruction tags corresponding to different paragraph structures to the lyrics.

[0006] In one possible design approach, after generating lyrics that carry visuals matching the musical feature data, the method further includes: Based on preset inspection rules, the lyrics are subjected to quality inspection to obtain quality inspection results; If the quality inspection results indicate that the quality has failed, the lyrics will be repaired in a targeted manner. If the quality check result indicates that the quality has passed, the lyrics are output.

[0007] Secondly, embodiments of this disclosure provide a lyrics generation apparatus, comprising: The user data acquisition module is used to acquire users' music feature data. The skeleton determination module is used to process the music feature data and determine the target skeleton template; the target skeleton template indicates the paragraph structure of the lyrics; The lyrics generation module is used to perform imagery anchoring processing based on the target skeleton template and the music feature data to generate lyrics that carry images that match the music feature data.

[0008] Thirdly, embodiments of this disclosure also provide an electronic device, including a memory storing multiple instructions; a processor loads instructions from the memory to execute the steps of any of the lyric generation methods provided in embodiments of this disclosure.

[0009] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform the steps of any of the lyric generation methods provided in embodiments of this disclosure.

[0010] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps in any of the lyric generation methods provided in embodiments of this disclosure.

[0011] The solution adopted in this embodiment first processes the music feature data to obtain a target skeleton template adapted to the user, achieving flexible matching of the skeleton template and avoiding severe homogenization of skeleton structures. Then, using the paragraph structure indicated by the target skeleton template as a fixed creative framework, and combining it with the user's music feature data for visual imagery anchoring, personalized lyrics tailored to the user are generated, satisfying the user's personalized creative needs. Furthermore, the lyrics content is driven by visuals, resulting in richer visuals and significantly enhancing emotional resonance and improving the user experience. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a lyrics generation method provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of a decision-making process provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of a creation process provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of a lyrics generation process provided in an embodiment of this disclosure; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this disclosure. Detailed Implementation

[0014] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure. Furthermore, in the description of the embodiments of this disclosure, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. Therefore, features defined with "first" or "second" may explicitly or implicitly include one or more features. In the description of the embodiments of this disclosure, "multiple" means two or more, unless otherwise explicitly specified.

[0015] The following is an explanation of the terms used in this disclosure.

[0016] Verse: The narrative section of a song (or music) used to explain the story, scene, emotional details, etc.

[0017] Chorus: The climax and memorable core of a song. The lyrics of the chorus are usually repeated.

[0018] Pre-Chorus: The transition section between the verse and the chorus of a song, gradually building up the emotion so that the chorus's climax is more impactful.

[0019] Bridge: Its melody, narrative content, and emotional atmosphere differ significantly from the verses and choruses. When a song has multiple choruses, the bridge generally follows the multiple choruses and precedes the last chorus. When a song has only one chorus, the bridge is located after the chorus and before the coda.

[0020] Coda: Also known as the outro, it is the final concluding section of a song. In this release, the coda can include a refrain coda and an instrumental outro. A refrain coda essentially reuses a short section of the chorus, retaining the main melody's memorable elements in the lyrics, emphasizing the narrative emotion at the end. An outro usually refers to a purely instrumental coda, but it can also include soft humming vocals, no longer using full lyrics, emphasizing the atmosphere of the arrangement at the end.

[0021] Interlude: This refers to a transitional section between different parts of a song. In this disclosure, interludes may include instrumental breaks and ambient sound breaks. Instrumental breaks primarily feature instrumental melodies, maintaining musical energy based on the groove and achieving a smooth transition between the preceding and following lyrical sections. Ambient breaks primarily use ambient sound effects and timbre, focusing on creating a narrative atmosphere and adjusting the emotional rhythm.

[0022] Beats per minute (BPM) indicates the tempo of a song.

[0023] In related technologies, most lyric generation methods adopt a single-stage generation mode of "theme → lyrics", which lacks in-depth profiling analysis of user behavior data and cannot realize personalized lyric creation based on the user's real musical personality, resulting in generated lyrics that cannot meet the user's personalized needs.

[0024] In addition, when using AI (artificial intelligence) large language models to directly generate lyrics, the following problems are likely to occur: (a) Abstraction of imagery: The extensive use of purely emotional words such as "missing", "loneliness" and "sadness" without concrete imagery results in the lack of appeal in the melodies generated by the AI ​​music engine that match the lyrics.

[0025] (b) Disorder of rhythm: The lyrics randomly generated by the AI ​​large language model often have chaotic rhymes and inconsistent rhymes within the same paragraph, which affects the memorability and melody fluency of the song.

[0026] (c) Flat emotional tone: The emotional intensity of each paragraph of the lyrics is similar, lacking a clear arc of introduction, development, transition and conclusion, resulting in a monotonous overall listening experience.

[0027] Therefore, to address the aforementioned issues, this disclosure provides a lyrics generation method that, based on multi-source user behavior data on music platforms, automatically generates lyrics text with highly personalized imagery density, strict rhythmic control, and a clear emotional arc through a multi-stage creation process driven by an AI agent. Furthermore, it achieves precise paragraph-level control over the downstream AI music generation engine through embedded metadata tags.

[0028] The implementation process of the above-mentioned lyrics generation method will be described in detail below. This implementation process can be performed by electronic devices (such as mobile phones, tablets, computers, servers, etc.). Figure 1 As shown, the implementation process of the lyrics generation method may include: S101. Obtain the user's music feature data.

[0029] Among them, music feature data includes at least one of music profile data, music location data, and music preference data.

[0030] For example, multi-source behavioral data of users on music platforms (such as listening to songs, favorites, playlists, playback duration, etc.) are analyzed to generate music feature data. This music feature data is structured data used to characterize users' long-term stable musical tastes and inner emotional needs, serving as the basis for lyric writing.

[0031] Optionally, music profile data (or deoxyribonucleic acid (DNA) profile data) may include multi-dimensional data from 10 dimensions: main language, core musical style, emotional preference, era preference, narrative preference, timbre preference, rhythm preference, instrument preference, life slice, and hidden personality. Each dimension of data contains a two-layer structure: an "information layer" (objective data anchors) and a "personality layer" (personality meaning inference).

[0032] Music location data (or user music identity positioning statement, or Identity Headline) generates a core personality definition statement based on a multi-candidate, multi-dimensional weighted method, providing a thematic anchor for lyric writing. For example, target feature data (such as music profile data) in the music feature data is assigned differentiated weights across multiple dimensions to calculate scores, outputting multiple sets of candidate personality descriptions. The statement with the highest weighted total score is selected as the core personality definition statement. This statement serves as the thematic anchor for the entire lyric, uniformly constraining the direction of image seed extraction, the overall emotional tone, and the narrative perspective of each paragraph, ensuring a high degree of match between the lyrical narrative and the user's inner personality.

[0033] The music preference data (or obsession insight library) includes obsession signals identified from multi-source behavioral data (such as obsession with a single artist, obsession with multiple versions, obsession across playlists, recent emotional outbursts, complete album acquisitions, version preferences, dual popularity / coldness, in-depth exploration of niche genres, timeline gaps, etc.) and Easter egg signals (such as nickname resonance, hidden rain in playlist naming, contrast signals, numerical destiny, etc.). In other words, the obsession insight data mines user scenarios, emotions, and recurring deep-seated obsession materials. Each obsession signal is accompanied by internal reasoning results for the obsession layer and lyric creation mapping suggestions. Optionally, the obsession layer includes at least layer 1, layer 2, and layer 3. Layer 1 represents a shallow obsession, characterized by objective reflection on events, no self-inflicted internal conflict, and an observer's perspective. Keywords are nostalgia, lamenting the passage of time, and reminiscing about scenery. For example, a mild obsession with multiple versions.

[0034] Level 2 represents intermediate-level obsession, characterized by the presence of emotional memories, but with psychological counseling completed and acceptance of reality achieved. Keywords: letting go, reconciling with the past, letting go, saying goodbye, and accepting reality.

[0035] Layer 3 represents deep-seated obsession, characterized by persistent inner turmoil, inability to reconcile, resentment, regret, repeated internal conflict, and intense tension. Keywords: unable to let go, remorse, struggle, confrontation, trapped in memories, and difficulty in reconciliation.

[0036] For example, user A's music profile data includes: emotional preference = sadness, narrative preference = reminiscence narrative, hidden personality = sealed-off, no-look-back personality, core music style = lyrical pop, rhythm preference BPM = 72, and instrument preference = piano.

[0037] User A's music location data: Prefers solitude on rainy nights and restrained, lyrical expressions of looking back on the past.

[0038] Music preference data (obsession insight library): Obsession layer number 3, deep obsession with three types of scene images: rainy night street lamp, blackboard, and bicycle in water.

[0039] Alternatively, the aforementioned music preference data can be identified from behavioral data using a 13-class signal detection algorithm.

[0040] Of course, the number of layers of obsession mentioned above (i.e., 3) is just an example. The number of layers can be configured according to the actual situation, as long as the higher the number of layers, the deeper the obsession.

[0041] Optionally, the overall emotional tone of the song can be determined based on emotional preferences, personality traits (or character traits), and obsession signals. For example, emotional preference labels, personality labels, and obsession signal labels can be mapped separately to obtain emotional preference scores, hidden personality scores, and obsession signal scores. Then, the emotional preference scores, hidden personality scores, and obsession signal scores are weighted and summed to obtain an emotional score. The corresponding emotional tone is determined based on the emotional score. Specifically, the target score interval to which the emotional partition belongs is found, and the emotional tone corresponding to that target score interval is used as the emotional tone corresponding to the emotional score. Different score intervals correspond to different emotional tones. For example, the emotional preference label: gentle and healing, has a mapping score S1 = 70. The hidden personality label: introverted and sensitive, has a mapping score of 60. The obsession signal label: letting go, reconciling with the past, has a mapping score S3 of 50. The emotional score is 70. 0.5+60 0.3+50 0.2.

[0042] Optionally, electronic devices can obtain music feature data through a user music profile building module.

[0043] In some embodiments, when a user wants to create audio (such as a song), the user can input lyric creation information, and the electronic device can generate corresponding musical skeleton features based on the lyric creation information and the user's musical feature data. Optionally, the lyric creation information may include style cue words, reference lyrics, etc.

[0044] For example, the electronic device can display interface 1, where the user can input lyric creation information into the relevant controls to trigger the electronic device to generate lyrics. Of course, the user can also input the information in other ways, such as by voice.

[0045] Alternatively, users can input audio information without directly entering it into their electronic devices. For example, the electronic device can act as a server, allowing the user to input lyrics on their phone. The phone then sends this lyrics to the server, triggering the server to generate lyrics. The server can then send the lyrics back to the phone.

[0046] S102. Process the music feature data to determine the target skeleton template. The target skeleton template indicates the paragraph structure of the lyrics.

[0047] The paragraph structure includes two layers of information: 1) paragraph type; 2) paragraph arrangement rules: number of verses, whether a pre-chorus is included, the position of the interlude, whether the ending uses a full chorus or a refrain, and the order of paragraphs. For details on the paragraph structure, please refer to the relevant descriptions below.

[0048] In some embodiments, different skeleton templates have corresponding skeleton features. The electronic device can first perform decision analysis on the music feature data to obtain music skeleton features. Then, it determines the skeleton template corresponding to the music skeleton features and uses it as the target skeleton template, thereby obtaining a skeleton template adapted to the user, instead of using a fixed skeleton template.

[0049] Optionally, the musical skeleton features represent structural constraints such as the number of sections, section structure, ending form, and special interlude configuration. Section structure represents the arrangement rules (i.e., the order in which section types are arranged) of section types. Section types include at least one of verse, chorus, bridge, pre-chorus, interlude, and coda. For example, an interlude may include ambient interludes and instrumental interludes. A coda may include a refrain coda and an instrumental coda.

[0050] In this embodiment of the disclosure, the paragraph structure of the lyrics is automatically determined based on the user's music feature data to obtain specific music skeleton features that fit the user's music needs.

[0051] In some embodiments, the analytical decision may include multi-level decision analysis. By performing multi-level decision analysis on a user's music feature data, music skeleton features adapted to the user's music features can be obtained. For example, the multi-level decision analysis process may include: determining basic skeleton features based on first data in the music feature data. The first data includes at least one of emotional preference, narrative preference, personality, and music preference data. Emotional preference and narrative preference belong to music profile data. Personality corresponds to the hidden personality (or hidden character) in the music profile data or the core personality (or core character) in the music positioning data.

[0052] Based on the second data in the music profile data, the basic skeleton features are verified and corrected to determine the music skeleton features. The second data includes rhythm preferences and / or core musical styles from the music profile data.

[0053] Then, the target skeleton template corresponding to the music skeleton features can be determined.

[0054] For example, the first data includes the core underlying data in the music feature data used to determine the basic paragraph arrangement. The second data represents the correction data in the music profile data used to adapt to the auditory style, including rhythm preference (BPM value), core musical style, etc.

[0055] In this embodiment, an agent performs a first-level decision (i.e., direct factor decision) based on first data, outputting basic skeletal features to determine the number of verses, pre-chorus configuration, ending form, whether to add an environment, and instrumental interludes. Then, the agent performs a second-level decision (i.e., indirect factor verification) based on second data, using the rhythm preferences and core musical style included in the second data as verification conditions to correct the basic skeletal features, eliminating conflicts between personality, narrative, emotional deduction, and musical style / rhythm compatibility, thus obtaining reasonable musical skeletal features.

[0056] For example, such as Figure 2 As shown, the first level of decision-making is direct factor decision-making. Basic structural features are determined based on personality, narrative preferences, and emotional preferences: the presence or absence of a pre-chorus, the number of verse sections, whether to add an interlude (such as an instrumental break), and whether to use a refrain to replace the full chorus at the end. It should be understood that "personality" here can refer to the hidden personality traits mentioned earlier or the personality traits corresponding to the Identity Headline.

[0057] Specifically, (a) Impulsive / Explosive Personality: No Pre-Chorus, Verse goes directly into the Chorus. (b) Gradual / Orderly Personality: Pre-Chorus is retained to create a tension progression from V→PC→C. (c) Sealed / No-Looking-back Personality: Refrain is used to replace the Final Chorus to achieve a low-pitched echo at the end. For example, the structure with the Final Chorus is V→PC→C→V→PC→C→Bridge→Final Chorus (Main Section) - Outro, while the structure with Refrain is V→PC→C→V→PC→C→Bridge→Refrain. (d) Number of layers of obsession ≥ 3: Three-section Verse structure. Number of layers of obsession ≤ 2: Two-section standard Verse structure. (e) Strong obsession with non-musical imagery: Add an Interlude section. (f) Extreme obsession with loops: Add an Instrumental Break section.

[0058] Optionally, music-related obsession signals refer to demands arising from the music itself, such as singers, albums, genres, versions, playlists, and rhythms (e.g., obsession with a single artist, complete album acquisition, version preference, etc.). Non-musical imagery obsessions are not within the music category and originate from the user's life, emotions, personal symbols, memory images, and numerical metaphors, such as nickname resonance, playlist naming cryptic messages, numerical destiny, and contrast signals mentioned above. Furthermore, strong non-musical imagery obsessions correspond to obsession layers greater than or equal to 3. It is understandable that the obsession keywords included in both non-musical imagery obsessions and musical imagery obsessions are preset. Accordingly, the agent determines whether a user's obsession falls under the category of obsession keywords included in non-musical imagery obsessions to determine whether the user's obsession is a non-musical imagery obsession.

[0059] The process of the first level of decision-making has been introduced above. The implementation process of the second level of decision-making will be introduced below.

[0060] The second layer of decision-making involves indirect factor verification. This utilizes rhythm preferences and core musical style to verify and correct the basic skeletal features output by the first layer of decision-making. For example, (a) fast tempo (BPM ≥ preset speed value 1, such as 120): if the basic skeletal features do not include a Pre-Chorus, then a short Pre-Chorus is added as an energy springboard, forming a V→PC→C tension progression.

[0061] (b) Specific music style adaptation rules: When the core music style is a preset style (such as R&B, Soul, etc.), it is necessary to rely on groove. If the basic skeleton features do not have a pre-Chorus, then a pre-Chorus is added to achieve a smooth transition. If the basic skeleton features have a pre-Chorus, then it is not necessary to add one.

[0062] (c) Conflict Handling Mechanism: When the BPM in the tempo preference is greater than or equal to 2 times the preset tempo value (e.g., 125) and there is no Pre-Chorus in the basic skeleton features, a Pre-Chorus is added. In cases where the section structure corresponding to the core style conflicts with the section structure corresponding to the hidden personality, the hidden personality is prioritized to determine the skeleton (i.e., the section structure corresponding to the hidden personality is retained). Additionally, in cases where there is a conflict between the traditional constraints of the style and the hidden personality, a corresponding arrangement tag can be generated based on the core style. This arrangement tag can be used as a structured production tag (or production instruction tag) to be encapsulated in a Metatag, achieving energy transition compensation through Metatag. For example, if the core style is R&B, its corresponding section structure is V-PC-C. The hidden personality's corresponding section structure is VC; therefore, VC is retained. At the boundary between the end of the Verse and the beginning of the Chorus, a Groove transition effect simulating the Pre-Chorus is added through arrangement techniques, thereby achieving energy buffering.

[0063] Optionally, the Pre-Chorus in the conflict resolution mechanism can be a short Pre-Chorus instead of a standard Pre-Chorus. The short Pre-Chorus is a lightweight form of the standard pre-chorus, and both have the same function (accumulating energy and smoothly transitioning the verse to the chorus), except that the duration of the short Pre-Chorus is shorter than that of the standard Pre-Chorus.

[0064] Optionally, adding a pre-chorus can be done by adding a pre-chorus between all verses and choruses (i.e., V and C) in the verse structure within the basic skeleton feature. Alternatively, a pre-chorus can be added between any or a specific V and C in the verse structure.

[0065] In this embodiment, a basic framework that aligns with the user's inner needs is first established using personality, emotions, and narrative to ensure emotional compatibility of the content. Then, a secondary verification based on musical style and rhythm preferences is introduced to avoid conflicts between the framework generated solely based on emotional personality and traditional musical style structures. This achieves a balance between inner emotional appeal and outer musical aesthetics, effectively preventing songs from having a structurally and emotionally compatible feel but a stylistically incongruous feel.

[0066] In some embodiments, after the second-level decision-making process, i.e., after verifying and correcting the basic skeleton features to obtain the corrected basic skeleton features, in one case, the corrected basic skeleton features are directly used as the music skeleton features. In another case, based on the instrument preferences in the music profile data, the instrumental data corresponding to the passages in the basic skeleton features are determined. The corrected basic skeleton features and the instrumental data are then superimposed to obtain the music skeleton features.

[0067] Among them, the instrumental data represents the instrumental passages or instruments corresponding to each passage (such as verse, pre-chorus, chorus, interlude, coda) assigned to the modified basic skeleton features.

[0068] In this embodiment of the disclosure, as described above Figure 2 As shown, multi-layer decision analysis also includes a third layer of decision-making: instrumental factor decision-making. The corrected basic skeleton features indicate the structure, order, length, transition methods, and ending forms of all sections in the entire song. Instrument preferences (i.e., instrument preference tags) are read from the DNA profile. For specific instrumental sections such as interludes within the section structure, or for each section individually, suitable playing instruments are matched to generate different corresponding playing instrument data. Then, the section structure information and the instrumental configuration information of each section are integrated and superimposed to form a complete musical skeleton feature. In addition, the playing instrument data can also be used to generate instrumental configuration tags, which serve as structured production tags, simultaneously guiding lyric generation and arrangement configuration. It can be understood that an instrumental section represents a section corresponding to a preset section type.

[0069] For example, if an acoustic guitar preference and / or minimalist arrangement preference are identified, the instrumental data indicates a fingerpicking acoustic guitar solo. If a piano preference and / or classical style preference are identified, the instrumental data indicates a piano, or a piano + strings interlude. If a traditional folk instrument preference is identified, the instrumental data indicates the corresponding folk instrument, such as a guzheng solo or an erhu solo. If no instrument preference exists, indicating no explicit instrument preference, the interlude content follows the main instrument selected by the style prompt (StylePrompt), meaning the instrumental data matches the style prompt. Optionally, the style prompt can be user-inputted or obtained by aggregating musical feature data. Optionally, the style prompt can correspond to a core personality in the Identity Headline.

[0070] By incorporating instrument preferences into the determination process of musical skeleton features, these features not only contain paragraph structure information but also carry the corresponding instrument configurations for each paragraph, thus achieving dual personalization of the song skeleton in both structural and timbre dimensions. Compared to methods that determine paragraph structure solely based on emotion and narrative, introducing instrument data allows skeleton features to match the user's aesthetic preferences for instruments, preventing the generated song skeleton from being out of sync with user preferences in terms of instrument configuration. Simultaneously, instrument data can also serve as the basis for generating structured production tags, providing paragraph-level instrument configuration guidance for downstream music generation engines and improving the collaborative accuracy between lyric generation and audio generation.

[0071] It should be noted that the skeletal characteristics corresponding to different personalities, narrative preferences, emotional preferences, obsessions, instrumental preferences, rhythm preferences, and core musical styles can be set according to actual needs, and this disclosure does not restrict them.

[0072] In some embodiments, after obtaining the music skeleton features through three layers of decision-making, the electronic device can determine the skeleton template that matches the music skeleton features from the preset skeleton templates and use it as the target skeleton template. For example, if the paragraph structure indicated by the music skeleton features is V→PC→C→V→PC→C→Bridge→Refrain, then the skeleton template corresponding to this paragraph structure is determined so that lyrics can be filled in different paragraphs in V→PC→C→V→PC→C→Bridge→Refrain using this skeleton template.

[0073] Optionally, the preset skeleton template includes at least one of the following: standard, narrative unfolding, pulse symmetry, energy progression, and free deconstruction. For example, if the musical skeleton features are Verse1→Pre-Chorus→Chorus→Interlude→Verse2→Pre-Chorus→Chorus→Bridge→Chorus (final chorus)→Outro, then a pulse symmetry skeleton template is matched.

[0074] For example, if the musical skeleton features are Verse1→Pre-Chorus→Chorus→Verse2→Pre-Chorus→Chorus→Bridge→Chorus→Outro, then it matches the standard skeleton template.

[0075] Optionally, the agent can also fine-tune the target skeleton template based on the music profile data. The specific fine-tuning process can be set according to requirements, and this disclosure does not limit it.

[0076] In some embodiments, the above three-layer decision-making can be performed by a three-layer structured decision module.

[0077] S103. Based on the target skeleton template, combine the music feature data to perform imagery anchoring processing and generate lyrics that carry images that match the music feature data.

[0078] In this embodiment of the disclosure, after determining the target skeleton template, the imagery anchoring method is used to generate lyrics that can reflect the scene. That is, the paragraph structure divided by the target skeleton template is used as a framework, and the user's music feature data is integrated to perform imagery anchoring processing, and the lyrics content of the concrete scene that fits the user's music needs is generated segment by segment.

[0079] In some embodiments, the implementation process of S104 above may include: extracting M picture seeds based on music profile data and music preference data; M is a positive integer. The M picture seeds are assigned to different paragraph structures. Lyrics are written based on the picture seeds corresponding to the paragraph structure to obtain the lyrics for that paragraph structure.

[0080] Among them, the image seed is defined as a micro-scene that can be perceived by sight, hearing, and touch.

[0081] In this embodiment of the disclosure, M image seeds are extracted from music profile data and music preference data based on image seed extraction rules. Then, the M (e.g., M = 6) image seeds are assigned to different paragraph structures (i.e., paragraphs) so that the image seeds corresponding to each paragraph can be used to drive the generation of lyrics for that paragraph. Notably, different paragraphs may contain the same image seeds.

[0082] Optionally, the image seed extraction rule can be at least one of the following rules: Extracting scene information based on listening history: Matching specific scenes based on songs that users repeatedly play on repeat, such as late at night, commuting, and rainy days; Extracting from music preference data: uncovering the specific behavioral perceptions behind obsessions, such as the auditory scenario of "only one song playing in the headphones" corresponding to the single-song loop behavior; Extracting environmental features based on slices of life (such as geographic and temporal signal data): Extracting environmental features with sensory recognition of the user's location, such as fog in city A or the sea in city B; Extract scenario keywords based on user-created playlist names.

[0083] Music feature data may also include the names of playlists created by users.

[0084] Optionally, the Agent can assign the M video seeds to different segments based on video seed usage rules. The video seed usage rules can be at least one of the following: The lyrics of the verse section must use at least P (e.g., 2) of image seeds; The chorus section must include at least R (e.g., 1) core picture seeds; Each segment must use at least T image seeds. In addition, the imagery of a segment can be abstracted, but all imagery must be generated from the extracted image seeds.

[0085] Optionally, when driving word writing, a list of purely emotional words can be set to be disabled. This list includes words such as longing, loneliness, sadness, happiness, heartache, sorrow, and bliss. These words cannot appear independently and can only be used to indirectly evoke the corresponding emotions through concrete imagery.

[0086] Optionally, when using a picture seed, the original picture seed can be used directly, or a picture seed derived from a variant of it can be used.

[0087] The core image seed is a distinctive image selected from all extracted image seeds, used to generate the Hook described below. For example, the image seed corresponding to the user's highest level of obsession is used as the core image seed. Alternatively, image seeds extracted from preset types of data (such as music preference data in the music feature data mentioned above) can be used as the core image seed.

[0088] In some embodiments, after writing lyrics based on image seeds, an image seed coverage scan can be performed on each paragraph of lyrics based on preset image density detection rules to achieve image density self-checking. The image anchoring method ends when each paragraph passes the check. If any paragraph fails the check, it indicates that the paragraph is unqualified, and targeted repair is performed.

[0089] The preset image density detection rules may include at least one of the following rules: The number of video seeds used in each verse section is greater than or equal to P; The number of core frame seeds used in each chorus section is greater than or equal to R; The number of purely emotional words in the lyrics is less than or equal to S (e.g., 1).

[0090] Optionally, the Agent can also output an image density self-check table for targeted repair of substandard paragraphs.

[0091] In this embodiment of the disclosure, based on the target skeleton template and music feature data, lyrics with matching visual content are generated using exclusive narrative scenes and imagery symbols, so that the lyrics have a visual quality.

[0092] In some embodiments, such as Figure 3 As shown, after determining the lyrics (here referred to as the initial lyrics) through the first creative process and the imagery anchoring method, the second creative process and the three-step rhyming method can be used to rhyme the initial lyrics. Based on the emotional tone of the lyrics, the rhyme schemes of the verse and chorus are determined. Based on the rhyme schemes of the verse and chorus, the lyrics are rhymed.

[0093] In this context, the verse rhyme scheme refers to the rhyme of the last word of the verse. For example, the verse rhyme scheme might be adapted to the narrative setup or a relatively calm mood of the verse. Optionally, the verse rhyme scheme may include one or more verse rhymes.

[0094] Similarly, the chorus rhyme scheme refers to the rhyme of the last word of the chorus verse. For example, the chorus rhyme scheme may be adapted to the rhyme scheme that releases the emotion of the chorus and reinforces the memory of the verse. Optionally, the verse rhyme scheme may include one or more verse rhymes.

[0095] The rhymes of the verse and the chorus are different, meaning there are no rhymes in the two, in order to create a contrast.

[0096] In this embodiment of the disclosure, the rhymes of the verse and chorus corresponding to the emotional tone of the entire song are determined. There is a mapping relationship between the emotional tone and the rhymes of the verse and chorus; different emotional tones correspond to different rhymes of the verse and chorus.

[0097] Then, based on the rhyme scheme of the verses, the lyrics of each verse section were adjusted so that the last word of each verse matched the rhyme scheme of the verses, thus unifying the ending rhymes of the verse sections while preserving the core imagery.

[0098] Based on the rhyme scheme of the chorus, the lyrics of each chorus section are adjusted so that the last word of each lyric matches the rhyme scheme of the chorus, thus unifying the ending rhyme of the chorus sections while preserving the core imagery.

[0099] In some embodiments, the Agent can select the chorus rhyme corresponding to the emotional tone of the entire song from a preset rhyme group (such as a preset 10 broad rhyme groups) according to preset rhyme selection rules. Furthermore, it can select rhymes from the remaining rhymes in the preset rhyme group (excluding the chorus rhyme) as the verse rhymes, thus creating a contrast between the verse and chorus rhymes.

[0100] After determining the rhymes for the verse and chorus, the agent can construct a rhyme dictionary. This dictionary can include a dictionary corresponding to the verse rhymes and a dictionary corresponding to the chorus rhymes. Optionally, the verse rhyme dictionary includes multiple words that rhyme with the verse rhymes. Similarly, the chorus rhyme dictionary includes multiple words that rhyme with the chorus rhymes.

[0101] For example, the above-mentioned preset rhyme selection rules include at least one of the following rules: Emotions like warmth or longing correspond to the rhyme "an"; Emotions of openness or relief correspond to the rhyme "ang"; Delicate or intimate emotions correspond to the i-rhyme group; Deep or profound emotions correspond to the "ong" rhyme group.

[0102] In some embodiments, the emotional tone includes at least one emotion, with different emotions corresponding to different rhymes, or one emotion corresponding to multiple rhymes. In one example, the agent can directly and randomly select one rhyme from multiple rhymes as the chorus rhyme.

[0103] In another example, the agent can use the rhymes corresponding to each emotion as candidate rhymes. Then, based on AI music engine characteristics (such as open vowel-ending rhymes (an, ang, ai, ao) being more conducive to melodic line extension than closed syllable rhymes (in, eng)), the agent selects the chorus rhyme from the candidate rhymes. For example, if the candidate rhymes include an and in, then an can be used as the chorus rhyme. Or, if the candidate rhymes include an, ang, and in, since an and ang are more suitable for melodic extension than in, the chorus rhyme can be selected from an and ang. Optionally, an or ang can be randomly selected from an and ang as the chorus rhyme. Alternatively, a priority can be set for the rhymes by default. The rhyme with the highest priority from an and ang is selected as the chorus rhyme.

[0104] Of course, there can be multiple rhymes in the chorus. However, usually, multiple chorus rhymes belong to the same chorus rhyme group (or broad rhyme group). A chorus rhyme group includes vowels that are auditoryally similar and that are allowed to rhyme interchangeably in the song. For example, "an" and "ang" belong to the same chorus rhyme group. Similarly, there can be multiple rhymes in the verse. However, usually, multiple verse rhymes belong to the same verse rhyme group.

[0105] In some embodiments, the process by which the Agent selects the verse rhyme from the remaining rhymes in a preset rhyme group (excluding the chorus rhyme) may include: the Agent randomly selecting one rhyme from the remaining rhymes as the verse rhyme. Alternatively, a priority is preset for the rhymes. The rhyme with the highest priority is selected from the remaining rhymes as the verse rhyme.

[0106] In some embodiments, the process of adjusting the lyrics of each verse section based on the rhyme scheme may include: for each verse section, adjusting the adjacent two lines or alternate lines of lyrics within that verse section, constrained by a dictionary corresponding to the rhyme scheme, so that the last character of the adjacent two lines or alternate lines of lyrics is a character from the dictionary corresponding to the rhyme scheme. That is, the verse section must at least use an AABB adjacent two-line rhyme format or an ABAB alternate-line rhyme format.

[0107] The process of adjusting the lyric lines of each chorus paragraph in the lyrics based on the chorus rhyme can include: for each chorus paragraph, taking the lexicon corresponding to the chorus rhyme as a constraint, adjusting each line of lyrics in the chorus paragraph, so that the last character of each line of lyrics is a character in the lexicon corresponding to the chorus rhyme. That is, the chorus at least adopts the full-line rhyme scheme. In addition, if there is a conflict in the narrative logic, priority is given to restructuring the sentence pattern to ensure that the established rhyme is not broken. For example, set the chorus rhyme: an, and its corresponding lexicon includes cold, hope, accompany, sigh. The lyrics determined based on the scene seed are: I wait alone for the dusk to come. But the rhyme of "come" does not correspond to an. Therefore, the sentence pattern can be restructured to: I wait alone for the twilight to turn cold.

[0108] Optionally, when adjusting the lyrics, the lyrics can be adjusted through a rhyme-driven method, for example, adjusting the last character rhyme first, and then retrofitting the entire line (i.e., the entire sentence).

[0109] In some embodiments, the Agent can not only perform rhyming processing on the lyrics of the two types of paragraphs, namely the verse and the chorus, but also perform rhyming processing on the lyrics of other types of paragraphs. Exemplarily, the lyrics of the pre-chorus paragraph are subjected to rhyming processing based on the pre-chorus rhyme. The process of determining the pre-chorus rhyme can refer to the above process of determining the chorus rhyme. Of course, the chorus rhyme can also be directly used as the pre-chorus rhyme. For example, adjust the last line of the Pre-Chorus lyrics so that the last character rhyme is the chorus rhyme, forming a natural transition.

[0110] In some embodiments, after generating lyrics (i.e., initial lyrics) based on the first creation process and the image anchoring method, the Agent can adjust the initial lyrics through the second creation process and the three-step rhyming method. The three-step rhyming method includes rhyme selection (i.e., selecting the verse rhyme and the chorus rhyme), rhyme lexicon-driven lyric writing (i.e., rhyming processing), and rhyme verification self-inspection. The rhyme selection and rhyme lexicon-driven lyric writing in the three-step rhyming method are introduced above. The rhyme verification self-inspection will be further introduced below.

[0111] After the rhyming processing is performed on the lyrics in the chorus paragraph, the lyrics after the rhyming processing can be verified line by line based on rhyming self-inspection rules (such as a rhyming self-inspection table). After the lyrics of each chorus paragraph pass the verification, it indicates that the lyrics after the rhyming processing meet the standards, and the lyrics after the rhyming processing can be directly output (or the lyrics after the rhyming processing are taken as target lyrics, and the target lyrics are output), or the lyrics after the rhyming processing can continue to be subjected to the related processing corresponding to the third creation process and the arc preset method.

[0112] If the verification of the lyrics of the chorus paragraph fails, which indicates that the lyrics after the rhyming processing do not meet the standards, the verse rhyme and the chorus rhyme are re-determined, so as to perform rhyming processing on the lyrics based on the new verse rhyme and chorus rhyme. Optionally, the emotional tone can also be re-determined.

[0113] Rhyme self-check rules may include at least one of the following rules: The rhyme of the last word of each line belongs to the rhyme group of the chorus. That is to say, the vowels contained in the rhyme group of the chorus are mutually recognized as valid rhymes. For example, an, ian, and uan can rhyme with each other. The number of near-rhyming words in a single chorus is less than or equal to 1 (e.g., 1). The number of times a single chorus has a rhyme scheme is less than or equal to 2 (e.g., 1).

[0114] Among them, quantity 1 and quantity 2 can be the same (e.g., both are 1) or different, and this disclosure does not restrict them.

[0115] Similar rhymes indicate words that don't belong to the chorus's rhyme group but have similar timbre. For example, if the chorus's rhyme group includes "an," the similar rhyme could be "en." Disjointed rhymes indicate words whose final rhyme doesn't belong to the chorus's rhyme group.

[0116] For example, the rhyme check rule includes that the number of similar rhymes in a single chorus is less than or equal to 1. If the number of similar rhymes in a single chorus is greater than 1, the check fails.

[0117] In this embodiment of the disclosure, the implementation of the three-step rhyming method achieves strict prosodic control of the entire line of Chorus, effectively solving the problem of chaotic rhymes when generating lyrics in large language models, and significantly improving the memorability and melody adaptability of the song.

[0118] As mentioned above, after completing the second creative process, the third creative process, the arc-preset method, can be executed. The process of the arc-preset method will be further described below. Of course, the agent can also skip the second creative process and directly execute the arc-preset method after executing the intention anchoring method. Correspondingly, the electronic device can differentiate the emotional intensity of each paragraph in the lyrics based on the emotional arc method to ensure that the lyrics have a clear emotional progression arc, with differentiation and progression between paragraphs. It should be noted that the lyrics here can be rhymed lyrics (or lyrics that have passed verification) or basic lyrics.

[0119] Among them, emotional intensity represents the level of emotional tension in a paragraph.

[0120] In some embodiments, the emotional arc method constructs a progressive emotional change curve for the entire song, assigning different emotional intensities according to paragraph type. For each paragraph in the lyrics, an arc card corresponding to that paragraph is generated based on musical feature data. The arc card includes the emotional intensity corresponding to the paragraph and emotional information. The emotional information includes at least one of emotional keywords, perspective, person, and narrative summary.

[0121] Based on the arc card corresponding to the paragraph, the lyrics of the paragraph are adjusted; the emotional intensity reflected by the adjusted lyrics matches the emotional intensity corresponding to the paragraph.

[0122] In this embodiment, the Agent determines the emotional tone based on music feature data. Then, based on the emotional tone, it assigns information such as emotional intensity (e.g., 1-5 stars), emotional keywords, perspective, person, and narrative summary to each paragraph, generating an arc card corresponding to each paragraph. Subsequently, the Agent uses the arc cards to drive adjustments to the lyrics, ensuring that the emotional intensity reflected in the adjusted lyrics matches the emotional intensity corresponding to that paragraph.

[0123] In some embodiments, the process of generating arc cards described above may include: Based on music feature data and combined with emotional constraints, arc cards corresponding to each paragraph are generated. The emotional constraints include at least one of the following conditions: The perspectives or point of view of adjacent verses are different; The emotional keywords in the bridge section are different from those in the chorus section; The emotional intensity of the ending is lower than that of the chorus; When the lyrics include N verses, the perspective or person of each of the N verses changes; N is a positive integer greater than 2. It should be understood that a change in the perspective of the N verses means that there are different perspectives among the N verses, and a change in person also means that there are different person perspectives.

[0124] By setting the aforementioned emotional constraints, homogenization of emotions and narrative perspectives across different verses can be effectively avoided. Specifically, the constraint of different perspectives or persons in adjacent verses differentiates the narrative angles of multiple verses, preventing the monotony caused by a single perspective in consecutive verses; the constraint of different emotional keywords in the bridge verses compared to the chorus ensures that the bridge verses provide new emotional information or narrative perspectives distinct from the chorus, preventing the bridge verses from becoming simple repetitions of the chorus's emotions; the constraint that the emotional intensity of the coda verses is lower than that of the chorus verses allows the song to naturally decline after its emotional climax, preventing a lack of closure due to excessively high emotions at the end; and the constraint of changing perspectives in multiple verses maintains narrative freshness when the song contains many verses, preventing listener fatigue.

[0125] For example, the target skeleton template carries a standard emotion arc. This standard emotion arc stores the basic emotion intensity corresponding to each paragraph. The agent uses the emotion tone, combined with the constraints on emotion intensity in the emotion constraints, to modify the basic emotion intensity of the paragraphs in the lyrics, thus obtaining the emotion intensity of that paragraph. For example, if the emotion tone is intense longing, the basic emotion intensity can be uniformly increased by 1, and then adjusted based on the constraints on emotion conditions.

[0126] For example, the target skeleton template carries a standard emotional arc. This standard emotional arc stores the basic emotional intensity corresponding to each paragraph. The agent uses the emotional tone, combined with the constraints on emotional intensity in the emotional constraints, to modify the basic emotional intensity of the paragraphs in the lyrics, thus obtaining the emotional intensity of that paragraph. For example, if the emotional tone is intense longing, the basic emotional intensity can be uniformly increased by 1, and then adjusted based on the constraints on emotional conditions. Afterwards, for each paragraph, based on the paragraph's emotional intensity and musical feature data, information such as emotional keywords, perspective, person, and narrative summary is assigned to the paragraph. Furthermore, this information needs to be constrained by the emotional constraints. For example, skeleton template: narrative unfolding type; emotional tone: relief. Final emotional intensity of each paragraph: V1=1, V2=2, PC=3, Chorus=5, Bridge=4, Outro=2; musical feature data: obsession level 2 (medium-level obsession, relief type), hidden introverted personality, narrative preference leaning towards reminiscence and introspection. Correspondingly, the emotional keyword determined based on the emotional intensity, musical feature data, and emotional constraints of V1 is "longing." The person is first-person. The perspective is immersive self-recollection. Narrative summary: Quietly recalling fragmented moments spent together. For example, the skeletal template is: pulse-symmetric type; the emotional tone is: intense longing. The final emotional intensity of each paragraph is: V1=2, PC=3, Chorus=5, Interlude=2, V2=3, PC=4, Chorus=5, Bridge=4, Chorus=5, Outro=3; musical feature data: Obsession level 3 (deep obsession, difficult to reconcile), hidden personality is pulse-burst type, narrative preference leans towards direct expression. Correspondingly, the emotional keyword determined based on the emotional intensity, musical feature data, and emotional constraints of V1 is "obsession." The person is second-person. The perspective is face-to-face questioning. Narrative summary: Directly expressing the emotional tug-of-war that cannot be let go.

[0127] In some embodiments, when the Agent adjusts lyrics using curved cards, for low emotional intensity paragraphs (i.e., paragraphs with emotional intensity less than the preset intensity), short sentences and descriptive imagery can be used, combined with emotional information, to adjust the lyrics of low emotional intensity paragraphs to make the tone of the text more gentle.

[0128] For high-emotion-intensity passages (i.e., passages with an emotional intensity greater than or equal to a preset intensity), at least one of the following methods can be used, in conjunction with emotional information, to adjust the lyrics of high-emotion-intensity passages: exclamatory sentences, word repetition, improvisational humming (Ad-lib), and high-density imagery. Additionally, if repeated perspectives are detected during the adjustment process (e.g., adjacent verse passages have the same perspective), the current lyrics adjustment process is terminated, and a new arc card is generated.

[0129] Among them, high-density imagery refers to the concentrated arrangement of multiple concrete visual symbols within a paragraph. By stacking images, emotional tension is enhanced, which is different from the method of depicting a single scene.

[0130] By employing an arc-shaped card mechanism, the emotional intensity of each paragraph is quantified and labeled, and structured emotional information such as emotional keywords, perspective, person, and narrative summary is added to form paragraph-level creative constraint cards. Compared to setting a global emotional tone for the entire song, arc-shaped cards provide an independent and quantifiable emotional control benchmark for each paragraph. This ensures that the emotional expression of each paragraph has a clear reference during lyric adjustment, thereby achieving precise emotional control at the paragraph level and avoiding the problem of unfocused emotional expression in lyrics caused by unclear emotional intensity in different paragraphs.

[0131] In some embodiments, after the lyrics are adjusted using the arc card driver, the adjusted lyrics are validated based on preset emotion validation rules. If the validation passes, the adjusted lyrics can be output (or the adjusted lyrics can be used as the target lyrics and output as the target lyrics). If the validation fails, the lyrics are readjusted based on the arc preset method. Optionally, adjusting the lyrics can refer to adjusting the lyrics of all paragraphs, or adjusting the lyrics of the paragraphs that failed the validation.

[0132] Optionally, the preset emotion verification rules include at least one of the following rules: The emotional intensity increases from the verse to the chorus to verify whether the transition from the verse to the chorus has a clear emotional rise and emotional outburst effect. Adjacent verses have different perspectives; these verses can be directly adjacent (e.g., VV) or indirectly identical (e.g., VCV).

[0133] The emotional keywords in the bridge section are different from those in the chorus section, in order to prevent simply replicating the emotions of the chorus.

[0134] The system uses preset emotion verification rules to check whether the intensity changes in the song's lyrics form an undulating arc, avoiding a flat emotional tone throughout. When the preset emotion verification rules include multiple rules, if multiple rules fail to verify, the verification is considered a failure. Otherwise, the verification is considered successful.

[0135] In some embodiments, the emotional intensity of a paragraph can be quantified. This is achieved by analyzing and weighting the emotional words, sentence structure features, and imagery density within the paragraph text. For example, if a paragraph contains words with strong positive emotions, the score corresponding to that word is used as the score for the emotional word.

[0136] In this embodiment, the arc preset method ensures that each section of the song has a clear emotional progression and perspective change, avoiding the common problem of "emotional flatness" in AI-generated lyrics, and giving the generated song the narrative tension of a professional lyricist.

[0137] In some embodiments, the above-mentioned arc preset method can be executed by the three major creative process modules.

[0138] In some embodiments, structure tags and production instruction tags corresponding to different paragraph structures are added to the lyrics. For example, such as... Figure 4 As shown, the Agent embeds paragraph-level metatags in the lyrics, enabling fine-grained control over the downstream AI music generation engine. The metatags use a combination of structural tags and production instruction tags, with each paragraph configured with a corresponding structural tag indicating its type.

[0139] Add production instruction tags after the structure tags (e.g., add production instruction tags in the format of hyphen + English), which indicate the production parameters such as arrangement, instrument configuration, vocal performance style, and dynamic energy corresponding to that section.

[0140] As introduced above, instruction tags can be determined based on music feature data. For example, when the obsession signal is a single artist's obsession, the vocal performance form corresponding to the segment is tagged as "A Cappella" to restore the pure vocal texture. When the obsession signal is a timeline discontinuity obsession, a "Silence" tag can be configured for segments (such as the epilogue), and the duration of the silence can also be marked to mark life milestones in the narrative.

[0141] In this embodiment of the disclosure, by embedding metatags in the lyrics text, precise collaborative control of the lyrics and the AI ​​music generation engine at the paragraph level is achieved. This makes the lyrics not only text content, but also precise instructions for the audio generation process, covering multiple dimensions such as orchestration, vocal performance, and dynamic energy, thereby achieving end-to-end AI music generation collaboration.

[0142] It should be noted that the lyrics mentioned above, which include structural tags and production instruction tags corresponding to different paragraph structures, can be the target lyrics obtained through creation process one, creation process two, or creation process three. It should be understood that if creation process one is executed instead of creation process two or three, then the lyrics obtained after executing creation process one can be used as the target lyrics.

[0143] In some embodiments, structural tags and production instruction tags corresponding to different paragraph structures can be added to the lyrics through the tag embedding module.

[0144] In some embodiments, the Agent can also perform quality checks on the generated lyrics (such as the target lyrics mentioned above). Based on preset inspection rules, the lyrics are inspected for quality, and the quality inspection results are obtained. If the quality inspection results indicate that the quality has failed, targeted repairs will be performed on the lyrics. If the quality check result indicates that the quality has passed, output the lyrics.

[0145] In this embodiment, the preset inspection rule includes at least one verification entry. The lyrics (e.g., target lyrics) are quality checked based on each verification entry, resulting in a quality inspection result for each entry. If a quality inspection result for a verification entry indicates that the quality has failed, targeted repair is performed on the lyrics based on that verification entry. If the instruction inspection results for all verification entries indicate that the quality has passed, the lyrics are output. Optionally, the lyrics may carry meta tags.

[0146] Optionally, the preset check rules may include at least one of the following check items: check items corresponding to structure and format (i.e., structure and format verification), check items corresponding to syllable consistency (i.e., syllable consistency verification), check items corresponding to hook quality (i.e., hook phrase quality verification), and check items corresponding to operation execution review (i.e., lyrics generation operation execution review). In other words, the preset check rules include at least one of the following verification dimensions: structure and format verification, syllable consistency verification, hook phrase quality verification, and lyrics generation operation execution review. A hook refers to a key short phrase in the lyrics that is highly recognizable, easy for listeners to remember, and carries the core theme. It is generally placed in the chorus area and is the core emotion and core imagery of the entire work. Simply put, a hook is the most catchy and memorable sentence in a song.

[0147] The validation items corresponding to the structure and format include at least one of the following: line ending punctuation, paragraph blank lines, paragraph line count, Ad-lib, ending mark, Style Prompt length, and meta tags, which must match the corresponding specification.

[0148] For example, line ending punctuation: Validate the punctuation at the end of each line of lyrics to ensure consistency and prevent redundant periods, commas, and question marks from interfering with the engine's text parsing.

[0149] Paragraph blank lines: Validate the blank line rules between different lyric paragraphs such as Verse, Chorus, and Bridge, and ensure consistent paragraph separation format.

[0150] Paragraph line count: Validates the number of lyrics lines in each paragraph (verse, chorus, bridge, etc.) to ensure they match the preset paragraph length of the corresponding type in the target skeleton template, preventing paragraphs from being too long or too short.

[0151] Ad-lib: Validates the marking format, occurrence position, and writing style of improvisational humming interjections, distinguishing between formal lyrics and improvisational vocal segments.

[0152] End marker: Verify the standard of the end marker for the entire song text to avoid text truncation and redundant content.

[0153] Style hint word length: Verifies whether the character length of the style hint word is less than or equal to the preset length to prevent the hint word from being too long and exceeding the input limit of the downstream engine.

[0154] Meta tags: Verify that the format of the meta tags embedded in the lyrics is correct, such as whether the tags are in English.

[0155] The check entries for syllable consistency may include at least one of the following: The character length difference between each line in the chorus section is less than or equal to the preset character length of 1 (e.g., 2 characters). The character length of each line in the chorus section is within the preset length range of 1 (e.g., 5 to 8 characters). The character length difference between each line in the verse is less than or equal to the preset character length of 2 (e.g., 3 characters). The character length of each line in the verse is within the preset length range of 2 (e.g., 6 to 8 characters); The character length of each line in the bridge paragraph is within the preset length range of 3 (e.g., 4 to 7 characters).

[0156] The verification entries corresponding to Hook quality may include at least one of the following: The hook position is in a preset position (such as the beginning or end of the chorus); The character length of the Hook must be less than or equal to the preset character length of 3 (e.g., 10 characters). Hooks include specific image words; among them, specific image words refer to words that carry visible and perceptible images, such as sunset, ferry, old letter, windowsill, starlight, withered leaves, alley lamp, and tide. A hook can stand alone, meaning it retains its complete meaning even without context and can function as a core slogan. For example, the lyric "The evening breeze takes away that year's dusk" is clear in its imagery and meaning even when taken alone, without needing other lyrics for comprehension. However, the lyric "Never again can wait" lacks a main body and, when extracted alone, may raise questions for users.

[0157] The verification items for operation execution review are as follows: The lyrics are obtained after performing all preset lyric generation operations. For example, it verifies that the imagery anchoring method, the three-step rhyme method, and the arc preset method have been executed. If the imagery anchoring method, the three-step rhyme method, or the arc preset method have not been executed, then the operation execution review is determined to have failed.

[0158] In some embodiments, the above-mentioned lyrics quality detection can be achieved through methods such as... Figure 4 The lyrics quality final inspection module shown is executed.

[0159] In this embodiment of the disclosure, the quality of lyrics is transformed from a subjective feeling of whether it is good or not into a standardized check that can be objectively measured through quality inspection, so as to ensure that each output lyric meets the preset quality threshold.

[0160] In some embodiments, such as Figure 5 As shown, the electronic device may include a user music profile building module, a structure decision module, a creation module, a tag embedding module, and a lyrics quality final inspection module.

[0161] The user music profile building module is used to determine the music characteristic data of the aforementioned users.

[0162] The structure decision module is used to determine the target skeleton template.

[0163] The creation module is used to generate personalized lyrics.

[0164] The lyrics quality final inspection module is used to check the quality of the lyrics.

[0165] The tag embedding module is used to add structural tags and production instruction tags corresponding to different paragraph structures in lyrics.

[0166] It should be noted that the acquisition and processing of user data involved in this disclosure are all done with the user's authorization and consent. Furthermore, the operations performed by the aforementioned Agent can be executed through a model or algorithm. For example, extracting a scene seed can be done by inputting relevant data into a relevant model, allowing the model to process the relevant data based on scene seed extraction rules to obtain the scene seed.

[0167] like Figure 6 As shown, Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. The electronic device 1100 includes a processor 1101 with one or more processing cores, a memory 1102 with one or more computer-readable storage media, and a computer program stored on the memory 1102 and executable on the processor. The processor 1101 and the memory 1102 are electrically connected. Those skilled in the art will understand that the electronic device structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0168] The processor 1101 is the control center of the electronic device 1100. It connects various parts of the electronic device 1100 via various interfaces and lines. By running or loading software programs and / or units stored in the memory 1102, and by calling data stored in the memory 1102, it executes various functions of the electronic device 1100 and processes data, thereby providing overall monitoring of the electronic device 1100. The processor 1101 can be a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), etc., and can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure.

[0169] In this embodiment of the disclosure, the processor 1101 in the electronic device 1100 loads the instructions corresponding to the processes of one or more applications into the memory 1102 according to the following steps, and the processor 1101 runs the applications stored in the memory 1102 to realize various functions, such as the steps of the lyrics generation method described above.

[0170] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0171] Optional, such as Figure 6 As shown, the electronic device 1100 also includes: a touch display screen 1103, a radio frequency circuit 1104, an audio circuit 1105, an input unit 1106, and a power supply 1107. The processor 1101 is electrically connected to the touch display screen 1103, the radio frequency circuit 1104, the audio circuit 1105, the input unit 1106, and the power supply 1107. Those skilled in the art will understand that... Figure 6 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0172] The touch display screen 1103 can be used to display a graphical user interface (GUI) and receive operation commands generated by the user interacting with the GUI. The touch display screen 1103 may include a display panel and a touch panel. The display panel can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of the electronic device. These graphical user interfaces can be composed of graphics, text, icons, video, and any combination thereof. Optionally, the display panel can be configured using a liquid crystal display (LCD), organic light-emitting diode (OLED), or other similar technologies. The touch panel can be used to collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel), generate corresponding operation commands, and execute the corresponding program according to the operation commands. Optionally, the touch panel may include a touch detection device and a touch controller. The touch detection device detects the user's touch location and the signal generated by the touch operation, transmitting the signal to the touch controller. The touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 1101. It can also receive and execute commands from the processor 1101. The touch panel can cover the display panel. When the touch panel detects a touch operation on or near it, it transmits the information to the processor 1101 to determine the type of touch event. Subsequently, the processor 1101 provides corresponding visual output on the display panel based on the type of touch event. In this embodiment, the touch panel and the display panel can be integrated into the touch display screen 1103 to achieve input and output functions. However, in some embodiments, the touch panel and the touch display screen 1103 can be implemented as two independent components to achieve input and output functions. That is, the touch display screen 1103 can also be used as part of the input unit 1106 to achieve input functions.

[0173] The radio frequency circuit 1104 can be used to transmit and receive radio frequency signals to establish wireless communication with network devices or other electronic devices, and to transmit and receive signals with network devices or other electronic devices.

[0174] Audio circuit 1105 can be used to provide an audio interface between a user and an electronic device via a speaker and a microphone. Audio circuit 1105 can convert received audio data into electrical signals and transmit them to the speaker, where the speaker converts them into sound signals for output. Conversely, the microphone converts collected sound signals into electrical signals, which are then received by audio circuit 1105, converted back into audio data, and then processed by processor 1101 before being transmitted via radio frequency circuit 1104 to, for example, another electronic device, or output to memory 1102 for further processing. Audio circuit 1105 may also include an earphone jack to provide communication between peripheral headphones and electronic devices.

[0175] The input unit 1106 can be used to receive input numbers, characters, or user characteristic information (such as fingerprints, iris, facial information, etc.), and to generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control.

[0176] Power supply 1107 is used to supply power to various components of electronic device 1100. Optionally, power supply 1107 can be logically connected to processor 1101 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. Power supply 1107 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0177] although Figure 6 As not shown in the diagram, the electronic device 1100 may also include a camera, sensor, wireless fidelity module, Bluetooth module, etc., which will not be described in detail here.

[0178] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0179] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0180] Therefore, embodiments of this disclosure provide a computer-readable storage medium storing a plurality of computer programs that can be loaded by a processor to execute any of the lyric generation methods provided in embodiments of this disclosure.

[0181] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0182] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0183] Since the computer program stored in the computer-readable storage medium can execute any of the virtual object recommendation methods provided in the embodiments of this disclosure, it can achieve the beneficial effects that any of the lyric generation methods provided in the embodiments of this disclosure can achieve, as detailed in the preceding embodiments, and will not be repeated here.

[0184] According to one aspect of this disclosure, a computer program product or computer program is also provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in the various optional implementations of the above embodiments.

[0185] In the above embodiments of computer-readable storage media, electronic devices, and computer program products, the descriptions of each embodiment have different focuses. Parts not described in detail in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes and beneficial effects of the computer-readable storage media, computer program products, electronic devices, and their corresponding units described above can be referred to the description of the virtual object recommendation method in the above embodiments, and will not be repeated here.

[0186] The foregoing has provided a detailed description of a lyric generation method, electronic device, and storage medium provided by the embodiments of this disclosure. Specific examples have been used to illustrate the principles and implementation methods of this disclosure. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this disclosure. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this disclosure. Therefore, the content of this specification should not be construed as a limitation of this disclosure.

Claims

1. A method for generating lyrics, characterized in that, include: Obtain user's music feature data; The music feature data is processed to determine the target skeleton template; The target skeleton template indicates the paragraph structure of the lyrics; Based on the target skeleton template, and combined with the music feature data, imagery anchoring processing is performed to generate lyrics that carry images that match the music feature data.

2. The method according to claim 1, characterized in that, After generating lyrics that carry images matching the music feature data by performing imagery anchoring processing based on the target skeleton template and the music feature data, the method further includes: Based on the emotional tone of the lyrics, determine the rhyme scheme of the verse and the rhyme scheme of the chorus; The lyrics are rhymed based on the rhymes of the verse and chorus.

3. The method according to claim 1 or 2, characterized in that, After generating lyrics that carry images matching the music feature data by performing imagery anchoring processing based on the target skeleton template and the music feature data, the method further includes: Based on the emotional arc method, the emotional intensity of each paragraph in the lyrics is adjusted differently.

4. The method according to claim 3, characterized in that, The method based on the emotional arc, which differentiates the emotional intensity of each paragraph in the lyrics, includes: For each paragraph in the lyrics, an arc card corresponding to the paragraph is generated based on the music feature data; wherein, the arc card includes the emotional intensity and emotional information corresponding to the paragraph; the emotional information includes at least one of emotional keywords, perspective, person, and narrative summary; Based on the arc card corresponding to the paragraph, the lyrics of the paragraph are adjusted; the emotional intensity reflected by the adjusted lyrics matches the emotional intensity corresponding to the paragraph.

5. The method according to claim 4, characterized in that, The step of generating the arc card corresponding to the paragraph based on the music feature data includes: Based on the music feature data and combined with emotional constraints, an arc card corresponding to the paragraph is generated; wherein the emotional constraints include at least one of the following conditions: The perspectives or point of view of adjacent verses are different; The emotional keywords in the bridge section are different from those in the chorus section; The emotional intensity of the ending is lower than that of the chorus; When the lyrics include N verses, the perspective or person of the N verses changes; N is a positive integer greater than 2.

6. The method according to claim 1 or 2, characterized in that, The music feature data includes music profile data and music preference data; The step of generating lyrics that carry images matching the music feature data based on the target skeleton template includes: M scene seeds are extracted based on the music profile data and music preference data; M is a positive integer. Assign M image seeds to different paragraph structures; The lyrics for the paragraph structure are obtained by driving the writing of words based on the image seed corresponding to the paragraph structure.

7. The method according to claim 1 or 2, characterized in that, The music feature data includes at least one of music profile data, music location data, and music preference data; The process of processing the music feature data to determine the target skeleton template includes: Based on the first data in the music feature data, basic skeleton features are determined; wherein, the first data includes at least one of emotional preference, narrative preference, personality and music preference data; the emotional preference and narrative preference belong to the music profile data; the personality corresponds to the hidden personality in the music profile data or the core personality in the music positioning data; Based on the second data in the music profile data, the basic skeleton features are verified and corrected to determine the music skeleton features; wherein, the second data includes rhythm preference and / or core musical style; Determine the target skeleton template corresponding to the music skeleton features.

8. The method according to claim 7, characterized in that, The step of verifying and correcting the basic skeleton features to determine the music skeleton features includes: The basic skeleton features are verified and corrected to obtain the corrected basic skeleton features. Based on the instrument preferences in the music profile data, determine the instrumental data corresponding to the segments in the basic skeleton features; The modified basic skeleton features and the instrumental data are superimposed to obtain the musical skeleton features.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing multiple instructions; the processor loads instructions from the memory to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the method as described in any one of claims 1 to 8.