AI-based music creation puzzle guessing system

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an AI-based music creation puzzle guessing system which comprises a data acquisition module, a music generation module, a music analysis module, a music processing module and a music optimization module. The data acquisition module is used for collecting music data of different years, regions and styles and input data of a user; the music generation module preprocesses music data, constructs a generation model and generates an initial music note sequence; the music analysis module analyzes the initial sequence extraction features and constructs a classification model to judge music attributes; the music processing module performs semantic analysis on input data, calculates similarity between a semantic matching result and a music attribute tag, and calculates a difficulty adjustment coefficient in combination with input time; and the music optimization module adjusts the initial difficulty level according to the difficulty adjustment coefficient, and further adjusts and generates model parameters to obtain a target model. The music creation difficulty can be dynamically adjusted according to the input data of the user, and personalized music guessing experience is achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI-based music creation and guessing system. Background Art

[0002] In today's digital entertainment era, music, as an important form of entertainment, is widely loved. With the rapid development of artificial intelligence (AI), its applications in various fields are expanding, including music. However, existing music entertainment methods and related products have many shortcomings, making it difficult to meet people's increasingly diverse and personalized needs.

[0003] Traditional music appreciation relies primarily on passive listening to existing works, lacking a deep sense of engagement and interaction with the music. While some music platforms offer simple interactive features like commenting and rating, these are relatively limited and prevent users from truly experiencing the creative process and essence of the music.

[0004] Professional music creation requires a high level of musical literacy and skills, making it difficult for ordinary users to participate. While some music creation tools for the general public have lowered the threshold for creation, their functionality is limited, and the quality of the music produced is often difficult to achieve. Furthermore, they lack social interactivity, making it difficult for user-generated works to be widely shared and exchanged.

[0005] Most existing music guessing games are based on existing classic music clips, asking players to guess song titles, singers, and other information. This type of game lacks innovation and fun, cannot fully leverage the advantages of artificial intelligence technology in music creation and analysis, and cannot meet users' needs for personalized music experience. Summary of the Invention

[0006] To this end, the present invention provides an AI-based music creation guessing system to solve the aforementioned problems existing in the prior art.

[0007] To achieve the above objectives, the present invention provides an AI-based music composition guessing system, comprising:

[0008] The data collection module is used to collect music data from different eras, regions, and styles, and also to obtain user input data and input time;

[0009] a music generation module, connected to the data acquisition module, for preprocessing the music data to obtain a processing result, and constructing a generation model based on the processing result and a preset initial difficulty level to generate different initial music note sequences;

[0010] a music analysis module, connected to the music generation module, configured to perform a multi-dimensional analysis on the initial music note sequence to extract feature results, and use the feature results as input to construct a classification model to determine the theme, emotion, and style of the music note sequence to obtain a music attribute label;

[0011] a music processing module, connected to the music analysis module, configured to perform semantic analysis on the input data to obtain a semantic result, match the semantic result with the music attribute label to obtain a similarity, and calculate a difficulty adjustment coefficient based on the similarity and the input time;

[0012] The music optimization module is connected to the music processing module, and adjusts the initial difficulty level according to the difficulty adjustment coefficient to obtain a target difficulty level, and adjusts the parameters of the generation model according to the target difficulty level to obtain a target generation model.

[0013] Furthermore, the music generation module includes:

[0014] a data processing unit, configured to perform data cleaning, feature extraction, and feature enhancement on the music data to obtain the processing results, wherein the processing results include event sequences, feature statistics, and feature embedding vectors;

[0015] a model building unit, connected to the data processing unit, for building the generative model based on the time series, the feature statistics and the feature embedding vector as input;

[0016] A sequence generation unit is connected to the model construction unit and is used to train the generation model according to the initial difficulty level to obtain different initial music note sequences.

[0017] Furthermore, the sequence generation unit includes:

[0018] a model training subunit, configured to train the generation model according to the initial difficulty level to obtain a training model;

[0019] a sequence generation subunit, connected to the model training subunit, for generating a musical note sequence according to the training model;

[0020] The difficulty adjustment subunit is connected to the sequence generation subunit and is used to adjust the complexity of the music note sequence according to the initial difficulty level to obtain the initial music note sequence.

[0021] Furthermore, the music analysis module includes:

[0022] a feature extraction unit for extracting features of the initial musical note sequence in time domain, frequency domain, rhythm, melody, harmony and structure to obtain the extraction result;

[0023] a feature selection unit connected to the feature extraction unit, for performing selection and dimensionality reduction processing on the extraction result to obtain a feature result;

[0024] a model training unit connected to the feature extraction unit, configured to use the feature result as input to construct a classification model, and train the classification model to determine the theme, emotion, and style of the initial musical note sequence to obtain a determination result;

[0025] The label generation unit is connected to the model training unit and is used to generate the music attribute label according to the judgment result.

[0026] Furthermore, the label generation unit includes:

[0027] a label mapping subunit, configured to map the judgment result to a predefined music attribute label vocabulary to obtain a mapping result;

[0028] The label optimization subunit is connected to the label mapping subunit and is used to optimize the mapping result to obtain the music attribute label.

[0029] Furthermore, the music processing module includes:

[0030] A semantic analysis unit, configured to perform word segmentation, tagging, and vectorization on the input data to obtain the semantic result;

[0031] a tag matching unit connected to the semantic analysis unit, for calculating the similarity between the semantic result and the music attribute tag using cosine similarity to obtain a calculation result;

[0032] A difficulty adjustment unit is connected to the tag matching unit and is used to calculate the difficulty adjustment coefficient according to the calculation result and the input time.

[0033] Furthermore, the semantic analysis unit includes:

[0034] A text segmentation subunit, configured to perform segmentation processing on the input data to obtain a segmentation result;

[0035] A text tagging subunit, connected to the text word segmentation subunit, for performing part-of-speech tagging on the word segmentation result to obtain a tagging result;

[0036] The vector generating subunit is connected to the text annotation subunit and is used to convert the annotation result into a semantic vector to obtain the semantic result.

[0037] Furthermore, the difficulty adjustment unit includes:

[0038] a similarity evaluation subunit, configured to compare the calculation result with a preset similarity range to obtain an evaluation result;

[0039] a time weighting subunit, configured to calculate a time factor according to a time decay function and the input time;

[0040] The coefficient calculation subunit is connected to the similarity evaluation subunit and the time weighting subunit respectively, and is used to calculate the difficulty adjustment coefficient according to the evaluation result and the time factor.

[0041] Furthermore, the music optimization module includes:

[0042] a difficulty level adjustment unit, configured to adjust the difficulty adjustment coefficient using a segmented adjustment method according to the initial difficulty level to obtain the target difficulty level;

[0043] a model parameter mapping unit connected to the difficulty level adjustment unit, for establishing a mapping relationship between the recognition result and the target difficulty level to obtain an adjustment parameter;

[0044] A model parameter adjustment unit is connected to the difficulty level adjustment unit and is used to adjust the generation model according to the adjustment parameters to obtain the target generation model.

[0045] Furthermore, the model parameter adjustment unit includes:

[0046] A parameter analysis subunit, configured to analyze the adjustment parameters and the adjustment range to obtain an analysis result;

[0047] a parameter positioning subunit, connected to the parameter parsing subunit, for locating the storage location of the parsing result according to the generation model to obtain a positioning result;

[0048] A parameter updating subunit is connected to the parameter parsing subunit and the parameter positioning subunit respectively, and is used to adjust the generation model according to the parsing result and the positioning result to obtain the target generation model.

[0049] Compared to the prior art, the present invention offers the advantage of obtaining user input data and input time, enabling the system to dynamically adjust based on user feedback and behavioral habits. By preprocessing music data and constructing a generative model, rapid and automated music creation is achieved, improving efficiency and lowering the barrier to entry. By constructing a generative model based on a preset initial difficulty level, initial musical note sequences of varying difficulty levels can be generated, providing a starting point for subsequent personalized adjustments. By performing multi-dimensional analysis on the initial musical note sequence, extracting feature results, and constructing a classification model to determine the theme, emotion, and style of the music, a deep understanding and precise classification of the music is achieved, providing a basis for subsequent matching and adjustment. By matching semantic results with music attribute labels and calculating a difficulty adjustment coefficient based on input time, accurate understanding of user intent and dynamic control of musical difficulty are achieved, enhancing the user experience. Dynamically adjusting the initial difficulty level and generative model parameters based on the difficulty adjustment coefficient enables personalized music customization, ensuring that the generated music better meets the user's preferences and needs. By continuously adjusting the generative model parameters, the system can continuously learn and optimize, continuously improving the quality of music creation and the fun of puzzle-solving.

[0050] In particular, the simulation training subunit trains the generative model based on the initial difficulty level, ensuring that the generative model can generate musical note sequences that meet the preset difficulty level at the initial stage. The sequence generation subunit generates musical note sequences based on the trained model, ensuring the consistency and coherence of the generated sequences. This helps improve the overall quality of the music, allowing users to experience a smoother and more enjoyable musical experience. By adjusting the complexity of the musical note sequence based on the initial difficulty level, the generated musical sequence can be better adapted to the musical proficiency and preferences of different users. This helps improve the personalization and adaptability of the system, allowing users to obtain a musical experience that better meets their expectations.

[0051] In particular, the label mapping subunit maps the judgment results obtained by the model training unit to a predefined vocabulary of music attribute labels, transforming abstract judgment results into specific music attribute labels. This makes complex musical features intuitive, allowing users to easily understand the musical connotations. The label optimization subunit further improves the accuracy and reliability of music attribute labels by optimizing the mapping results. Through optimization methods such as threshold processing, label fusion, label sorting, and conflict resolution, redundant and conflicting labels can be eliminated, ensuring that the final set of music attribute labels is more refined and accurate.

[0052] In particular, the text segmentation subunit can better understand the user's input intention by breaking down the text data input by the user into smaller, meaningful vocabulary units. This helps to accurately capture the user's music preferences and needs, and provide more refined input information for subsequent music processing and generation. The text annotation subunit can further understand the grammatical function and semantic role of words in sentences by tagging the word segmentation results. This helps to gain a deeper understanding of the user's input data. The vector generation subunit can convert text information into numerical form by converting the annotation results into semantic vectors, which helps to effectively match the user's input data with music attribute labels, providing more accurate input for the music creation guessing system.

[0053] In particular, the similarity assessment subunit precisely controls the similarity of generated content, ensuring that the music adheres to the intended style and theme while maintaining a certain level of innovation. The time-weighted subunit introduces a time decay mechanism, allowing the difficulty adjustment to dynamically change over the generation process, avoiding premature restrictions or late-stage rigidity and allowing the model to adopt different generation strategies at different stages. The coefficient calculation subunit calculates the adjustment coefficient by combining similarity and time factors, achieving more intelligent and detailed control of generation difficulty.

[0054] In particular, the similarity assessment subunit precisely controls the similarity of generated content, ensuring that the music adheres to the intended style and theme while maintaining a certain level of innovation. The time-weighted subunit introduces a time decay mechanism, allowing the difficulty adjustment to dynamically change over the generation process, avoiding premature restrictions or late-stage rigidity and allowing the model to adopt different generation strategies at different stages. The coefficient calculation subunit calculates the adjustment coefficient by combining similarity and time factors, achieving more intelligent and detailed control of generation difficulty.

[0055] In particular, the parameter parsing subunit accurately locates the parameters requiring adjustment within the model by parsing parameter names or indices, avoiding adjustments to incorrect or irrelevant parameters. The parameter positioning subunit analyzes the model structure to accurately locate the storage location of the parameters requiring adjustment, preventing parameter adjustment failures or model anomalies caused by positioning errors. The parameter updating subunit flexibly adjusts model parameters based on different adjustment requirements, enabling precise control over the music generation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of the structure of an AI-based music composition and guessing system provided in an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of the structure of an AI-based music composition and guessing system provided in an embodiment of the present invention;

[0058] Figure 3A schematic diagram of the structure of an AI-based music composition and guessing system provided in an embodiment of the present invention;

[0059] Figure 4 A schematic diagram of the structure of an AI-based music composition guessing system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0062] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0063] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0064] See also Figure 1 As shown, the AI-based music creation and guessing system provided by the embodiment of the present invention includes:

[0065] The data collection module 10 is used to collect music data of different eras, regions and styles, and also to obtain the user's input data and input time;

[0066] Specifically, public music resources and metadata (age, region, style tags, etc.) are obtained from major online music platforms (such as Spotify, Apple Music, NetEase Cloud Music, QQ Music, etc.) through API interfaces or web crawlers. Input data generated during the interaction between users and the system is collected, especially the answers users give to the generated music. The types of data collected include the results of users' guesses about the generated music, which can be text descriptions (such as song title, artist, age, style, etc.), multiple-choice options, or other forms. User behavior data during the guessing process is recorded, such as the length of time the song is listened to, the number of guesses, the time it takes to choose the answer, etc. User feedback on the guessing results and the system as a whole is collected. The time when users input data is collected to provide a time dimension reference for subsequent difficulty adjustments and personalized recommendations.

[0067] a music generation module 20 connected to the data acquisition module 10, configured to pre-process the music data to obtain a processing result, and to construct a generation model based on the processing result and a preset initial difficulty level to generate different initial music note sequences;

[0068] a music analysis module 30 connected to the music generation module 20 for performing a multi-dimensional analysis on the initial music note sequence to extract feature results, and using the feature results as input to construct a classification model to determine the theme, emotion, and style of the music note sequence to obtain a music attribute label;

[0069] a music processing module 40 connected to the music analysis module 30, configured to perform semantic analysis on the input data to obtain a semantic result, match the semantic result with the music attribute label to obtain a similarity, and calculate a difficulty adjustment coefficient based on the similarity and the input time;

[0070] The music optimization module 50 is connected to the music processing module 40, and adjusts the initial difficulty level according to the difficulty adjustment coefficient to obtain a target difficulty level, and adjusts the parameters of the generation model according to the target difficulty level to obtain a target generation model.

[0071] Specifically, by acquiring user input data and input time, the system can dynamically adjust based on user feedback and behavioral habits. Preprocessing music data and constructing a generative model enables rapid and automated music creation, improving efficiency and lowering the barrier to entry. Building a generative model based on a preset initial difficulty level generates initial musical note sequences of varying difficulty levels, providing a starting point for subsequent personalized adjustments. By performing multi-dimensional analysis on the initial musical note sequence, extracting feature results, and constructing a classification model to determine the theme, emotion, and style of the music, the system achieves in-depth understanding and accurate classification of the music, providing a basis for subsequent matching and adjustment. By matching semantic results with music attribute labels and calculating a difficulty adjustment coefficient based on input time, the system accurately understands user intent and dynamically controls the difficulty of the music, enhancing the user experience. Dynamically adjusting the initial difficulty level and generative model parameters based on the difficulty adjustment coefficient enables personalized music customization, ensuring that the generated music better suits the user's preferences and needs. By continuously adjusting the generative model parameters, the system continuously learns and optimizes, continuously improving the quality of the music creation and the fun of the puzzle.

[0072] Specifically, if Figure 2 As shown, the music generation module 20 includes:

[0073] a data processing unit 21 for performing data cleaning, feature extraction, and feature enhancement on the music data to obtain the processing results, wherein the processing results include event sequences, feature statistics, and feature embedding vectors;

[0074] a model building unit 22, connected to the data processing unit 21, for building the generative model according to the time series, the feature statistics and the feature embedding vector as input;

[0075] The sequence generation unit 23 is connected to the model construction unit 22 and is used to train the generation model according to the initial difficulty level to obtain different initial music note sequences.

[0076] Specifically, the data processing unit first cleans the input music data to remove noise, outliers, and irrelevant information. For example, it might remove silent segments from the music data or correct incorrectly recorded notes. Next, it extracts useful features from the cleaned music data. These features may include: converting the music data into a series of events, such as the start and end times of notes and their pitch; calculating statistics such as the average pitch, average duration, and range of notes; and using embedding techniques to convert the music data into a vector representation in a high-dimensional space to better capture the patterns and structure within the music data. After extracting features, this subunit may also enhance the features to improve model performance. For example, it may use data augmentation techniques to generate more training data or feature selection techniques to select the most relevant features. The model building unit uses this input data to construct a generative model. The generative model can be a deep learning model, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), or a Transformer. These models are capable of learning sequential patterns and structures in music data and generating new music sequences. During the model construction process, this subunit can also configure model parameters as needed, such as the number of layers, number of hidden units, activation function, etc. The sequence generation unit trains the generative model based on the initial difficulty level. The initial difficulty level can be a numerical value or a classification label, indicating the initial difficulty of the music sequence. During training, this subunit uses training algorithms, such as gradient descent, to adjust the parameters of the generative model so that it can generate music sequences that meet the initial difficulty level. Finally, this subunit uses the trained generative model to generate different initial musical note sequences. These sequences can serve as starting points for music creation or be used for other music-related tasks.

[0077] Specifically, the data processing unit removes noise and irrelevant information through data cleaning to ensure the accuracy of subsequent processing and modeling. The model building unit uses the time series, feature statistics, and feature embedding vectors output by the processing unit to construct a generative model, enabling the model to accurately capture the patterns and structures in the music data. The sequence generation unit trains the generative model based on the initial difficulty level, generating music sequences that meet different difficulty levels to meet diverse musical needs.

[0078] Specifically, the sequence generation unit includes:

[0079] a model training subunit, configured to train the generation model according to the initial difficulty level to obtain a training model;

[0080] a sequence generation subunit, connected to the model training subunit, for generating a musical note sequence according to the training model;

[0081] The difficulty adjustment subunit is connected to the sequence generation subunit and is used to adjust the complexity of the music note sequence according to the initial difficulty level to obtain the initial music note sequence.

[0082] Specifically, in the simulation training subunit, if labeled data for different difficulty levels is available, a data subset matching or approximating the initial difficulty level can be selected for training. Alternatively, data augmentation techniques (such as altering note duration, rhythmic density, and harmonic complexity) can be used to simulate training data of a specific difficulty level. The generative model's loss function is modified to incorporate difficulty-related constraints. For example, a penalty term based on music theory metrics (such as average note density, chord change frequency, interval span, and rhythmic complexity) can be added to the loss function. For low difficulty levels, the penalty weight for sequences with high complexity can be increased; for high difficulty levels, the penalty weight can be reduced or a reward term encouraging complexity can be added. Difficulty levels are used to guide the model in learning specific musical styles or structures, which are often associated with a certain level of difficulty. Using the mapped training objectives / constraints, the input generative model is further trained (fine-tuned). The training process is similar to traditional model training, including forward propagation, loss calculation, backpropagation, and parameter updates. The goal of training is to ensure that the model's output characteristics (such as rhythm, harmony, and melodic direction) when generating music are more likely to meet the requirements of the initial difficulty level. The sequence generation unit prepares the initial conditions required for the generation process, such as the starting event (which may be silence, a simple chord, or a random note), a context vector (if required by the model), a random seed, etc. The trained model is configured for generation mode. The current sequence (or its representation) is fed into the model. The model predicts the next event (or the probability distribution of the next event). The next specific event is sampled from the predicted probability distribution. For example, a temperature parameter is used to adjust the sharpness of the probability distribution, and then sampling is performed according to the probability. Note: The temperature parameter here may be affected by the difficulty implicit in the training model, but it mainly controls the randomness of the generation process rather than directly adjusting the complexity. More direct difficulty control occurs in the difficulty adjustment subunit. The sampled event is added to the generated sequence. The current sequence is updated, and the above process is repeated until the preset generation length is reached, the end marker is encountered, or other stopping conditions are met. The difficulty adjustment subunit evaluates the complexity of the musical note sequence output by the sequence generation subunit. A combination of various music theory indicators can be used, such as rhythmic complexity (average note duration, note density, rhythmic change frequency, presence of syncopation, etc.), melodic complexity (interval span (especially large jumps), changes in melodic direction, use of ornaments, etc.), harmonic complexity (number of chords, chord change frequency, chord type (triad, seventh chord, etc.), chord inversion, etc.) and overall statistics (average pitch, dynamic range and variation, etc.). The calculated current complexity value is compared with the target complexity range or threshold corresponding to the initial difficulty level. Based on the comparison results, the generated music note sequence is modified so that its complexity approaches the target difficulty: Reduce complexity (if the current complexity > target): merge short notes into long notes, remove syncopation, and unify rhythmic patterns.Narrow the intervals, use a smoother melodic progression, and remove ornaments. Reduce the number of chords, use simpler chords (such as triads instead of sevenths), and lengthen chord durations. Reduce the dynamic range. Increase complexity (if current complexity is less than target): insert shorter notes, add syncopation, and change the rhythmic pattern. Use a wider interval, increase the twists and turns of the melodic progression, and add ornaments. Increase the number of chords, use more complex chords, shorten chord durations, and increase chord inversions. Introduce richer dynamics.

[0083] Specifically, the simulation training subunit trains the generative model based on the initial difficulty level, ensuring that the generative model can generate musical note sequences that meet the preset difficulty level in the initial stage. The sequence generation subunit generates musical note sequences based on the trained model, ensuring that the generated sequences are consistent and coherent. This helps improve the overall quality of the music, allowing users to have a smoother and more enjoyable musical experience. By adjusting the complexity of the musical note sequence based on the initial difficulty level, the generated music sequence can be better adapted to the musical levels and preferences of different users. This helps to improve the personalization and adaptability of the system, allowing users to obtain a musical experience that better meets their expectations.

[0084] Specifically, the music analysis module includes:

[0085] a feature extraction unit for extracting features of the initial musical note sequence in time domain, frequency domain, rhythm, melody, harmony and structure to obtain the extraction result;

[0086] a feature selection unit connected to the feature extraction unit, for performing selection and dimensionality reduction processing on the extraction result to obtain a feature result;

[0087] a model training unit connected to the feature extraction unit, configured to use the feature result as input to construct a classification model, and train the classification model to determine the theme, emotion, and style of the initial musical note sequence to obtain a determination result;

[0088] The label generation unit is connected to the model training unit and is used to generate the music attribute label according to the judgment result.

[0089] Specifically, the feature extraction unit analyzes the temporal variations of musical note sequences, extracting features such as note onset, duration, and volume. Using methods such as Fourier transform, the time-domain signal is converted to the frequency domain to extract features such as frequency content, spectral centroid, and spectral bandwidth. The rhythmic pattern of the note sequence is analyzed to extract features such as beat, rhythmic pattern, and rhythmic complexity. The melodic direction of the note sequence is analyzed to extract features such as pitch, interval, and melodic contour. The harmonic structure of the note sequence is analyzed to extract features such as chord type, chord progression, and harmonic complexity. The overall musical structure is analyzed, including features such as form, paragraph division, and thematic repetition. From the feature extraction results, a subset of features with strong correlations with musical attributes (theme, emotion, and style) is selected. Methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) are used to reduce the feature dimensionality and improve the model's computational efficiency and generalization capabilities. Appropriate machine learning algorithms, such as support vector machines (SVMs), random forests (RFs), and neural networks (NNs), are selected to construct a classification model. The classification model is trained using labeled music data, and model performance is optimized by adjusting model parameters. The trained classification model is then used to classify the feature results of the initial musical note sequence, generating judgments regarding theme, emotion, and style. Based on the judgments obtained during the model training unit, corresponding musical attribute labels are generated for the initial musical note sequence, such as "classical music," "cheerful emotion," or "complex style." The generated musical attribute labels are then mapped to specific musical attributes for subsequent music processing and optimization.

[0090] Specifically, the feature extraction unit extracts features from the initial musical note sequence in the time domain, frequency domain, rhythm, melody, harmony, and structure, comprehensively capturing all key musical information. This helps to gain a deeper understanding of the music's intrinsic properties. The feature selection unit selects and reduces the dimensionality of the extracted results, removing redundant and irrelevant features while retaining features highly correlated with the musical attributes. This helps improve the model's computational efficiency and generalization capabilities, avoiding overfitting and the curse of dimensionality. The model training unit uses the feature results as input to construct and train a classification model, learning the mapping relationship between musical features and musical attributes. This helps accurately determine the theme, emotion, and style of the initial musical note sequence, providing a reliable basis for subsequent music processing and optimization. The label generation unit generates music attribute labels based on the judgment results obtained by the model training unit, which can intuitively represent the various musical attributes. This helps users understand and select music and also facilitates subsequent music retrieval and recommendation.

[0091] Specifically, the label generation unit includes:

[0092] a label mapping subunit, configured to map the judgment result to a predefined music attribute label vocabulary to obtain a mapping result;

[0093] The label optimization subunit is connected to the label mapping subunit and is used to optimize the mapping result to obtain the music attribute label.

[0094] Specifically, the label mapping subunit creates a vocabulary containing various music attribute labels, such as "classical music," "pop music," "jazz," "cheerful emotion," "sad emotion," "complex style," and "simple style." The judgment results obtained by the model training unit (such as the probability distribution of theme, emotion, and style) are mapped to the predefined vocabulary of music attribute labels. This can be achieved through a lookup table or conditional probability table, associating the probability value of each attribute with the most relevant label. A preliminary set of music attribute labels is generated, with each label corresponding to the probability value of an attribute. The label optimization subunit sets a probability threshold. Only when the probability value of an attribute exceeds this threshold is the corresponding label included in the final set of music attribute labels. When multiple labels have high probability values, it is possible to consider fusing these labels into a more general label, for example, fusing "pop music" and "rock music" into "pop / rock music." Labels are sorted by probability, with the label with the highest probability value placed first, giving it higher priority when displayed or used. When there are conflicts or overlaps between labels, they are resolved according to specific rules or algorithms to ensure the accuracy and consistency of the final label set. After optimization, a final set of music attribute labels is generated, which will be used to describe the music attributes of the initial music note sequence.

[0095] Specifically, the label mapping subunit maps the judgment results obtained by the model training unit to a predefined vocabulary of music attribute labels, transforming abstract judgment results into specific music attribute labels. This makes complex musical features intuitive, allowing users to easily understand the musical connotations. The label optimization subunit further improves the accuracy and reliability of music attribute labels by optimizing the mapping results. Through optimization methods such as threshold processing, label fusion, label sorting, and conflict resolution, redundant and conflicting labels can be eliminated, ensuring that the final set of music attribute labels is more refined and accurate.

[0096] Specifically, if Figure 3 As shown, the music processing module 40 includes:

[0097] A semantic analysis unit 41, configured to perform word segmentation, tagging, and vectorization on the input data to obtain the semantic result;

[0098] a tag matching unit 42 connected to the semantic analysis unit 41 and configured to calculate the similarity between the semantic result and the music attribute tag using cosine similarity to obtain a calculation result;

[0099] The difficulty adjustment unit 43 is connected to the tag matching unit 42 and is used to calculate the difficulty adjustment coefficient according to the calculation result and the input time.

[0100] Specifically, the label matching unit calculates cosine similarity between the vector output by the semantic analysis unit and the vector of each music attribute label. Cosine similarity measures the cosine of the angle between two vectors in vector space, and its value range is [-1, 1]. Larger values ​​indicate more similarity between the two vectors, meaning that the semantic result is more relevant to the label. For example, if the theme of the music is "tranquil forest" and the player answers "quiet woods", the similarity calculated by semantic analysis may be around 0.8; if the player answers "bustling city", the similarity may be close to 0.

[0101] Specifically, the label matching unit uses cosine similarity to calculate the similarity between the semantic results and the music attribute labels, which can quantify the degree of correlation between the semantic results and the music attribute labels. This helps select the music attribute labels that are most relevant to the user input, improving the accuracy and pertinence of music generation.

[0102] Specifically, the semantic analysis unit includes:

[0103] A text segmentation subunit, configured to perform segmentation processing on the input data to obtain a segmentation result;

[0104] A text tagging subunit, connected to the text word segmentation subunit, for performing part-of-speech tagging on the word segmentation result to obtain a tagging result;

[0105] The vector generating subunit is connected to the text annotation subunit and is used to convert the annotation result into a semantic vector to obtain the semantic result.

[0106] Specifically, the text tokenization subunit uses a predefined dictionary to match the longest words in the text. For example, for the input "I like listening to music", the dictionary contains words such as "like", "listen", and "music", and the tokenization result is "I / like / listen / music". Use a statistical language model to perform tokenization based on the co-occurrence frequency of words. For example, for the input "I like listening to music", the statistical model may find that "like listen" often appears together, so the tokenization result may be "I / like listen / music". Use a neural network model, such as a recurrent neural network (RNN) or a long short-term memory network (LSTM), to perform tokenization. For example, a pre-trained BERT model can be used for tokenization to obtain a more accurate tokenization result. The text annotation subunit uses predefined grammar rules and dictionaries to perform词性标注 (POS tagging) on words. For example, according to the rule that "a noun is usually followed by a verb", "I / like / listen / music" can be annotated as "I / rr / like / v / listen / v / music / n" (rr: pronoun, v: verb, n: noun). Output the annotation result (the word sequence and its corresponding POS tags). The vector generation subunit represents the text as a set of words, ignoring word order and grammatical information. For example, for the annotation result "I / rr like / v listen / v music / n", it can be represented as a vector, where each dimension corresponds to a word and the dimension value is the number of times the word appears. Based on the bag-of-words model, considering the discriminability of words, downweight common words. For example, for the word "of", since its occurrence frequency is very high but its discriminability is very low, its TF-IDF value will be low. Use a neural network model to map words to vectors in a high-dimensional space so that semantically similar words are close in the vector space. For example, a pre-trained Word2Vec model can be used to map "like" and "love" to similar vectors. Use a pre-trained BERT model to encode the text into semantic vectors.

[0107] Specifically, by decomposing the text data input by the user into smaller and meaningful lexical units, the text tokenization subunit can better understand the input intention of the user. This helps to accurately capture the user's music preferences and needs, providing more refined input information for subsequent music processing and generation. By performing POS tagging on the tokenization result, the text annotation subunit can further understand the grammatical role and semantic role of words in a sentence. This helps to more deeply understand the user's input data. By converting the annotation result into semantic vectors, the vector generation subunit can transform the text information into a numerical form, which helps to effectively match the user's input data with music attribute tags, providing more accurate input for the music creation guessing system.

[0108] Specifically, the difficulty adjustment unit includes:

[0109] It should be noted that the term "词性标注" in the original text is not a standard English expression. I have translated it as "POS tagging" which is a common abbreviation in the field of natural language processing. If there is a more specific English term for this concept in your context, it can be adjusted accordingly.A similarity evaluation subunit, which is used to compare the calculation result with a preset similarity range to obtain an evaluation result;

[0110] A time weighting subunit, which is used to calculate a time factor according to a time decay function and the input time;

[0111] A coefficient calculation subunit, which is respectively connected to the affiliated similarity evaluation subunit and the time weighting subunit, and is used to calculate the difficulty adjustment coefficient according to the evaluation result and the time factor.

[0112] Specifically, the similarity evaluation subunit inputs the similarity score obtained from the label matching unit. Internally stores or receives a set of preset similarity range thresholds. These thresholds can define different similarity levels, for example: a low similarity threshold (e.g., 0.3), a medium similarity threshold (e.g., 0.7), and a high similarity threshold (e.g., 0.9). Compare the received similarity score with the preset threshold: If the similarity score is less than the low similarity threshold: the evaluation result is low similarity. If the similarity score is between the low similarity threshold and the medium similarity threshold: the evaluation result is medium similarity. If the similarity score is greater than the high similarity threshold: the evaluation result is high similarity. The time weighting subunit internally defines a time decay function. For example, a simple linear decay or exponential decay function can be used. According to the input time and the time decay function, calculate the time factor. The time factor represents the degree of influence of the input time on the difficulty adjustment coefficient. The coefficient calculation subunit obtains the evaluation result from the similarity evaluation subunit. Obtains the time factor from the time weighting subunit. According to the evaluation result and the time factor, calculates the difficulty adjustment coefficient. The specific calculation method is as follows:

[0113] C = B(S) * [1 / (1 + (T / τ))]

[0114] Where B(S) indicates that B is a value determined according to the comparison of the similarity score S with the preset threshold, S is the similarity score, T is the input time, τ is the time constant, B is the base adjustment coefficient, C is the final difficulty adjustment coefficient, and F is the time factor.

[0115] The formula for calculating the time factor is: F = 1 / (1 + (T / τ)). This formula indicates that as time increases, the value of F will gradually decrease. The value of B depends on the comparison of S with the preset similarity threshold:

[0116] If S < Threshold_Low (low similarity threshold), then B = Value_Low (low - level similarity base adjustment coefficient, for example, 0.5);

[0117] If Threshold_Low < S < Threshold_High (high similarity threshold), then B = Value_Medium (medium similarity base adjustment coefficient, e.g., 1.0);

[0118] If S > Threshold_High, then B = Value_High (high similarity base adjustment coefficient, e.g., 1.5).

[0119] Specifically, the similarity evaluation subunit ensures that the music conforms to the expected style and theme while maintaining a certain degree of innovation by precisely controlling the similarity of the generated content. The time-weighting subunit introduces a time decay mechanism, enabling the difficulty adjustment to change dynamically during the generation process, avoiding premature restrictions or late-stage rigidities, and allowing the model to have different generation strategies at different stages. The coefficient calculation subunit calculates the adjustment coefficient by comprehensively considering the similarity and time factors, achieving a more intelligent and meticulous regulation of the generation difficulty.

[0120] Specifically, as Figure 4 shown, the music optimization module 50 includes:

[0121] A difficulty level adjustment unit 51 for adjusting the difficulty adjustment coefficient using a piecewise adjustment method according to the initial difficulty level to obtain the target difficulty level;

[0122] A model parameter mapping unit 52 connected to the difficulty level adjustment unit 用51 to establish a mapping relationship between the recognition result and the target difficulty level to obtain adjustment parameters;

[0123] A model parameter adjustment unit 53 connected to the difficulty level adjustment unit 52 for adjusting the generation model according to the adjustment parameters to obtain the target generation model.

[0124] Specifically, difficulty level adjustment divides the difficulty level into several intervals (e.g., easy, medium, and hard), each corresponding to a different adjustment strategy. A set of adjustment rules is pre-defined for each interval. For example, for the "easy" interval, the difficulty adjustment coefficient may need to be reduced to reduce the complexity of the generated music; for the "hard" interval, the difficulty adjustment coefficient may need to be increased to increase the complexity of the generated music. Based on the initial difficulty level and the corresponding adjustment strategy, the difficulty adjustment coefficient is calculated to obtain the final target difficulty level. For example, if the initial difficulty level is "medium," the corresponding adjustment strategy is to multiply the difficulty adjustment coefficient by 0.8, and the target difficulty level is the adjusted coefficient. The model parameter mapping unit analyzes the recognition results and extracts key features. For example, if the recognition result is "classical music," it may be necessary to extract features related to classical music, such as melodic complexity and harmonic richness. A mapping relationship is established between the target difficulty level and the key features of the recognition result. This can be achieved using machine learning algorithms (such as regression and classification). For example, a model can be trained with the target difficulty level and the features of the recognition result as input, and the output is the parameters used to adjust the generation model. Using the established mapping relationship, specific adjustment parameters are calculated based on the current target difficulty level and recognition results. These parameters are used to guide the model parameter adjustment unit in adjusting the generative model. The model parameter adjustment unit adjusts the internal parameters of the generative model based on the adjustment parameters output by the model parameter mapping unit. This may include adjusting the weights of the neural network, changing the probability distribution of the generation rules, and so on. Different adjustment strategies are adopted depending on the adjustment parameters. For example, if the adjustment parameters indicate a need to reduce the complexity of the generated music, the generation probability of certain notes or chords can be reduced. After parameter adjustment, the resulting target generative model will be able to generate music according to the target difficulty level.

[0125] Specifically, segmented adjustment and parameter mapping enable finer control over the generative model, enabling targeted adjustments based on varying difficulty levels and musical characteristics. The music optimization module can adaptively adjust based on different recognition results and target difficulty levels, enhancing the system's flexibility and versatility. By optimizing the generative model, music pieces can be generated that better match the target difficulty level and musical characteristics, improving the quality of the generated music and the user experience.

[0126] Specifically, the model parameter adjustment unit includes:

[0127] A parameter analysis subunit, configured to analyze the adjustment parameters and the adjustment range to obtain an analysis result;

[0128] a parameter positioning subunit, connected to the parameter parsing subunit, for locating the storage location of the parsing result according to the generation model to obtain a positioning result;

[0129] A parameter updating subunit is connected to the parameter parsing subunit and the parameter positioning subunit respectively, and is used to adjust the generation model according to the parsing result and the positioning result to obtain the target generation model.

[0130] Specifically, the parameter parsing subunit identifies each parameter's name or index (indicating the parameter in the model that needs to be adjusted) and the corresponding adjustment amount. Adjustment amounts can be of two types: Increment: This is a specific numeric value indicating how much the parameter's current value should be increased or decreased. For example, +0.5 indicates an increase of 0.5, while -0.3 indicates a decrease of 0.3. Multiplier: This is a scaling factor indicating how much the parameter's current value should be multiplied. For example, *1.2 indicates a 20% increase in the parameter's value, while *0.8 indicates a 20% decrease. After parsing is complete, the subunit generates an internal data structure that clearly records each parameter's name / index, the adjustment type (increment or multiplier), and the specific adjustment value. This data structure is passed to the next subunit. The parameter location subunit first receives the parsing result from the parameter parsing subunit, which contains the name / index of the parameter to be adjusted. Based on the parameter name / index in the parsed result, the subunit searches within the generated model to locate each parameter object and its precise location within the model. This typically involves traversing and accessing the model structure. After positioning is complete, the subunit generates a positioning result data structure that clearly records each parameter object and its location in the model. The parameter update subunit receives the parsing results (including the adjustment type and value) from the parameter parsing subunit and the positioning results (including the parameter object and its location) from the parameter positioning subunit. Based on the positioning results, the parameter object and its location in the model are found. Based on the adjustment type and value in the parsing results, the specific adjustment operation is performed: If it is an incremental adjustment, the current value of the parameter object is added or subtracted from the incremental value in the parsing result. For example, if the parsing result is +0.5, the parameter value is increased by 0.5. If it is a multiplicative adjustment, the current value of the parameter object is multiplied by the multiplier value in the parsing result. For example, if the parsing result is *1.2, the parameter value is increased by 20%. The adjusted parameter values ​​are written back to the corresponding locations in the model. This ensures that after all parameters are adjusted, the internal state of the model is updated to reflect the new parameter values.

[0131] Specifically, the parameter parsing subunit parses parameter names or indices to accurately locate the parameters that need adjustment in the model, avoiding adjusting incorrect or irrelevant parameters. The parameter positioning subunit analyzes the model structure to accurately locate the storage location of the parameters that need adjustment, avoiding parameter adjustment failures or model anomalies caused by positioning errors. The parameter updating subunit can flexibly adjust model parameters according to different adjustment requirements, thereby achieving precise control over the music generation process.

[0132] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0133] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An AI-based music creation guessing system, characterized by: include: The data collection module is used to collect music data from different eras, regions, and styles, and also to obtain user input data and input time; a music generation module, connected to the data acquisition module, for preprocessing the music data to obtain a processing result, and constructing a generation model based on the processing result and a preset initial difficulty level to generate different initial music note sequences; a music analysis module, connected to the music generation module, configured to perform a multi-dimensional analysis on the initial music note sequence to extract feature results, and use the feature results as input to construct a classification model to determine the theme, emotion, and style of the music note sequence to obtain a music attribute label; a music processing module, connected to the music analysis module, configured to perform semantic analysis on the input data to obtain a semantic result, match the semantic result with the music attribute label to obtain a similarity, and calculate a difficulty adjustment coefficient based on the similarity and the input time; The music optimization module is connected to the music processing module, and adjusts the initial difficulty level according to the difficulty adjustment coefficient to obtain a target difficulty level, and adjusts the parameters of the generation model according to the target difficulty level to obtain a target generation model.

2. The AI-based music composition guessing system according to claim 1, characterized in that: The music generation module includes: a data processing unit, configured to perform data cleaning, feature extraction, and feature enhancement on the music data to obtain the processing results, wherein the processing results include event sequences, feature statistics, and feature embedding vectors; a model building unit, connected to the data processing unit, for building the generative model based on the time series, the feature statistics and the feature embedding vector as input; A sequence generation unit is connected to the model construction unit and is used to train the generation model according to the initial difficulty level to obtain different initial music note sequences.

3. The AI-based music composition guessing system according to claim 2, characterized in that: The sequence generation unit comprises: a model training subunit, configured to train the generation model according to the initial difficulty level to obtain a training model; a sequence generation subunit, connected to the model training subunit, for generating a musical note sequence according to the training model; The difficulty adjustment subunit is connected to the sequence generation subunit and is used to adjust the complexity of the music note sequence according to the initial difficulty level to obtain the initial music note sequence.

4. The AI-based music composition guessing system according to claim 3, characterized in that: The music analysis module includes: a feature extraction unit for extracting features of the initial musical note sequence in time domain, frequency domain, rhythm, melody, harmony and structure to obtain the extraction result; a feature selection unit connected to the feature extraction unit, for performing selection and dimensionality reduction processing on the extraction result to obtain a feature result; a model training unit connected to the feature extraction unit, configured to use the feature result as input to construct a classification model, and train the classification model to determine the theme, emotion, and style of the initial musical note sequence to obtain a determination result; The label generation unit is connected to the model training unit and is used to generate the music attribute label according to the judgment result.

5. The AI-based music composition guessing system according to claim 4, characterized in that: The label generation unit includes: a label mapping subunit, configured to map the judgment result to a predefined music attribute label vocabulary to obtain a mapping result; The label optimization subunit is connected to the label mapping subunit and is used to optimize the mapping result to obtain the music attribute label.

6. The AI-based music composition guessing system according to claim 5, characterized in that: The music processing module includes: A semantic analysis unit, configured to perform word segmentation, tagging, and vectorization on the input data to obtain the semantic result; a tag matching unit connected to the semantic analysis unit, for calculating the similarity between the semantic result and the music attribute tag using cosine similarity to obtain a calculation result; A difficulty adjustment unit is connected to the tag matching unit and is used to calculate the difficulty adjustment coefficient according to the calculation result and the input time.

7. The AI-based music composition guessing system according to claim 6, characterized in that: The semantic analysis unit includes: A text segmentation subunit, configured to perform segmentation processing on the input data to obtain a segmentation result; A text tagging subunit, connected to the text word segmentation subunit, for performing part-of-speech tagging on the word segmentation result to obtain a tagging result; The vector generating subunit is connected to the text annotation subunit and is used to convert the annotation result into a semantic vector to obtain the semantic result.

8. The AI-based music composition guessing system according to claim 7, characterized in that: The difficulty adjustment unit includes: a similarity evaluation subunit, configured to compare the calculation result with a preset similarity range to obtain an evaluation result; a time weighting subunit, configured to calculate a time factor according to a time decay function and the input time; The coefficient calculation subunit is connected to the similarity evaluation subunit and the time weighting subunit respectively, and is used to calculate the difficulty adjustment coefficient according to the evaluation result and the time factor.

9. The AI-based music composition guessing system according to claim 8, characterized in that: The music optimization module includes: a difficulty level adjustment unit, configured to adjust the difficulty adjustment coefficient using a segmented adjustment method according to the initial difficulty level to obtain the target difficulty level; a model parameter mapping unit connected to the difficulty level adjustment unit, for establishing a mapping relationship between the recognition result and the target difficulty level to obtain an adjustment parameter; A model parameter adjustment unit is connected to the difficulty level adjustment unit and is used to adjust the generation model according to the adjustment parameters to obtain the target generation model.

10. The AI-based music composition guessing system according to claim 9, characterized in that: The model parameter adjustment unit includes: A parameter analysis subunit, configured to analyze the adjustment parameters and the adjustment range to obtain an analysis result; a parameter positioning subunit, connected to the parameter parsing subunit, for locating the storage location of the parsing result according to the generation model to obtain a positioning result; A parameter updating subunit is connected to the parameter parsing subunit and the parameter positioning subunit respectively, and is used to adjust the generation model according to the parsing result and the positioning result to obtain the target generation model.

Images