Song scoring model training method and song scoring method

Abstract

The application relates to a song scoring model training method and a song scoring method. The method comprises the following steps: obtaining song audio, comments of a first sample song, and comments of a second sample song; inputting the song audio into an audio feature extraction model to obtain song audio features, and inputting the comments of the first sample song and the comments of the second sample song into a comment feature extraction model to obtain comment features of the first sample song and comment features of the second sample song; and training a song scoring model according to the similarity between the song audio features and the comment features of the first sample song and the similarity between the song audio features and the comment features of the second sample song. The method can analyze the correlation between the comments and the song audio from the perspective of the comments to score the song, so that the song scoring is not limited to pure audio analysis, the song is scored more flexibly by fully utilizing the comment information, and the accuracy of the song scoring is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a song scoring model training method and song scoring method, apparatus, computer equipment, storage medium and computer program product. Background Technology

[0002] Singing evaluation involves analyzing the vocal audio of the person being evaluated, considering dimensions such as timbre, pitch, rhythm, breath control, vibrato, glissando, and emotion, to arrive at a comprehensive score. This score allows the person being evaluated to understand their strengths and weaknesses, increasing the enjoyment of karaoke. For example, when different users sing the same song, each user's performance can be analyzed to determine their score, thus identifying the winner.

[0003] Traditional techniques train relevant models by scoring songs and then use the trained models to determine the song scores. However, traditional techniques only score songs by analyzing and learning the relationship between songs and their corresponding specific scores. When the scores cannot fully cover all segments, traditional techniques cannot learn the differences between different segments. Furthermore, song scores based on different scoring standards can lead to inaccuracies in the final song scores, which is not conducive to improving the accuracy of song scoring. Summary of the Invention

[0004] Therefore, it is necessary to provide a song scoring method, apparatus, computer device, computer-readable storage medium, and computer program product that can improve the accuracy of song scoring in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for training a song rating model, wherein the song rating model includes an audio feature extraction model and a comment feature extraction model, comprising:

[0006] Obtain the audio information of the first sample song and the comment information of the first sample song, and obtain the comment information of the second sample song that is different from the first sample song;

[0007] The song audio information is input into the audio feature extraction model to be trained to obtain the song audio features of the first sample song. The comment information of the first sample song and the comment information of the second sample song are input into the comment feature extraction model to be trained to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song.

[0008] Based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, a song rating model to be trained is trained, resulting in a trained song rating model.

[0009] In one embodiment, the comment information of the first sample song includes a first type of comment information and a second type of comment information; the comment information of the second sample song includes the first type of comment information and the second type of comment information; the first type of comment information is obtained by the user through input from the music platform, and the second type of comment information is obtained based on expert experience;

[0010] The step of training a song rating model based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, includes:

[0011] The song audio information of the first sample song, the first type of comment information of the first sample song, and the first type of comment information of the second sample song are input into the song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the first type of comment features of the first sample song based on the first type of comment information of the first sample song, and the first type of comment features of the second sample song based on the first type of comment information of the second sample song.

[0012] The song audio information of the first sample song, the second type of comment information of the first sample song, and the second type of comment information of the second sample song are input into the song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the second type of comment features of the first sample song based on the second type of comment information of the first sample song, and the second type of comment features of the second sample song based on the second type of comment information of the second sample song.

[0013] The song rating model to be trained is trained based on at least the song audio features of the first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song.

[0014] In one embodiment, training the song rating model to be trained based at least on the song audio features of the first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song includes:

[0015] The song rating model to be trained is trained based on the first similarity between the song audio features of the first sample song and the first type of comment features of the first sample song, the second similarity between the song audio features of the first sample song and the first type of comment features of the second sample song, the third similarity between the song audio features of the first sample song and the second type of comment features of the first sample song, and the fourth similarity between the song audio features of the first sample song and the second type of comment features of the second sample song.

[0016] In one embodiment, training the song rating model to be trained based at least on the song audio features of the first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song includes:

[0017] Based on the first similarity between the audio features of the first sample song and the first type of comment features of the first sample song, and the second similarity between the audio features of the first sample song and the first type of comment features of the second sample song, the song scoring model to be trained is trained to obtain a preliminary trained song scoring model.

[0018] The preliminarily trained song rating model is trained based on the third similarity between the song audio features of the first sample song and the second type of comment features of the first sample song, and the fourth similarity between the song audio features of the first sample song and the second type of comment features of the second sample song.

[0019] Secondly, this application also provides a song scoring method, including:

[0020] The song to be scored is input into a pre-trained song scoring model to extract the song's audio features; wherein the pre-trained song scoring model is trained by the song scoring model training method.

[0021] In a pre-defined comment database, at least one comment whose comment features meet a pre-defined similarity condition with the song audio features of the song to be scored is identified as the target comment; wherein, the comment features of the comment information in the comment database are obtained by the pre-trained song scoring model;

[0022] Based on the rating information corresponding to each of the target comment information, the song rating of the song to be rated is determined.

[0023] In one embodiment, the step of inputting the song to be scored into a pre-trained song scoring model to extract the song's audio features includes:

[0024] The feature mapping network in the audio feature extraction model of the song scoring model maps the song audio information of the song to be scored to a preset space to obtain the mapping features corresponding to the song audio information.

[0025] The feature transformation network in the audio feature extraction model converts the mapping features corresponding to the song audio information into the song audio features of the song to be scored.

[0026] In one embodiment, determining the song rating of the song to be rated based on the rating information corresponding to each of the target comment information includes:

[0027] Based on the weight information corresponding to each of the target comment information, the score information corresponding to each of the target comment information is weighted and summed to obtain the song score of the song to be scored.

[0028] Thirdly, this application also provides a song rating model training device, wherein the song rating model includes an audio feature extraction model and a comment feature extraction model, comprising:

[0029] The acquisition module is used to acquire the audio information of the first sample song, the comment information of the first sample song, and the comment information of the second sample song that is different from the first sample song.

[0030] The extraction module is used to input the song audio information into the audio feature extraction model to be trained to obtain the song audio features of the first sample song, and to input the comment information of the first sample song and the comment information of the second sample song into the comment feature extraction model to be trained, so as to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song.

[0031] The training module is used to train the song rating model to be trained based on the similarity between the song audio features of the first sample song and the comment features of the first sample song, and the similarity between the song audio features of the first sample song and the comment features of the second sample song, so as to obtain the trained song rating model.

[0032] Fourthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the steps of the method described above.

[0033] Fifthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0034] Sixthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0035] The aforementioned song scoring model training method, song scoring method, device, computer equipment, storage medium, and computer program product utilize the audio information of sample songs and the corresponding comments to train the song scoring model. This allows the song scoring model to learn the correlation between song audio and corresponding comments. Based on the pre-trained song scoring model, it accurately selects target comments corresponding to the songs to be scored. It then accurately selects at least one target comment from the comment database that meets a preset similarity condition with the audio features of the song to be scored. Combining the scoring information corresponding to each target comment, it determines the song score for the song to be scored. Therefore, the embodiments of this application can analyze the correlation between song comments and song audio from the perspective of song comments to score songs. This allows song scoring to move beyond pure audio analysis and eliminates the need for a strict score quantification mechanism. By fully utilizing comment information, it enables more flexible and varied scoring of songs, thereby improving the accuracy of song scoring. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a diagram illustrating the application environment of a song scoring model training method in one embodiment.

[0038] Figure 2 This is a flowchart illustrating a song scoring model training method in one embodiment;

[0039] Figure 3 This is a flowchart illustrating a song scoring method based on a cross-modal pre-trained large model in one embodiment;

[0040] Figure 4 This is a structural block diagram of a song scoring model training device in one embodiment;

[0041] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0043] The song scoring method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Server 104 obtains the audio information of a first sample song and its comment information, and obtains the comment information of a second sample song that is different from the first sample song. Server 104 inputs the song audio information into an audio feature extraction model to be trained to obtain the audio features of the first sample song, and inputs the comment information of the first and second sample songs into a comment feature extraction model to be trained, so as to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song. Server 104 trains a song rating model to be trained based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, to obtain the trained song rating model. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0044] In one exemplary embodiment, such as Figure 2 As shown, a method for training a song rating model is provided. The song rating model includes an audio feature extraction model and a comment feature extraction model. The method is illustrated using a server as an example. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0045] Step S202: Obtain the audio information of the first sample song and the comment information of the first sample song, and obtain the comment information of the second sample song that is different from the first sample song.

[0046] The first sample song can refer to any song on the music platform. In practical applications, the first sample song can include songs on the music platform that have corresponding user comments and expert comments.

[0047] The comment information for the first sample song can refer to information characterizing the performance of the first sample song. In practical applications, the comment information for the first sample song can include user comments entered by users on the music platform and expert comments based on expert experience. User comments in the comment information for the first sample song can be obtained from the comment section of the first sample song on the music platform, while expert comments can be obtained from music rating personnel (such as musicians). It is understood that the form of comment on the song can include, but is not limited to, text and audio formats.

[0048] The second sample song can refer to a song on the music platform that is different from the first sample song.

[0049] The comment information for the second sample song can refer to information that characterizes the evaluation of the performance of the second sample song.

[0050] As an example, to make the song rating model output more objective and accurate, the server can train the song rating model. Specifically, the training samples of the song rating model to be trained can include a sample song dataset in the music platform. The sample song dataset includes several sample songs. The server can use the audio information of the first sample song, the comment information of the first sample song, and the comment information of the second sample song that is different from the first sample song as sample data to train the song rating model.

[0051] Step S204: Input the song audio information into the audio feature extraction model to be trained to obtain the song audio features of the first sample song, and input the comment information of the first sample song and the comment information of the second sample song into the comment feature extraction model to be trained to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song.

[0052] The audio feature extraction model refers to a model used to determine the audio features of a song to be scored. In practical applications, the audio feature extraction model can encode the audio information of the song to be scored to obtain the audio features of the song. The audio feature extraction model can include an audio representation model. After the audio representation model, an MLP (Multilayer Perceptron) module or other neural networks used for data preprocessing, feature extraction, and representation learning can be concatenated to form a review feature extraction model. The audio representation model can include, but is not limited to, the Mert model (a music analysis model) and the Hubert model (a speech analysis model). The MLP module can adjust the representation dimension of the output result of the audio representation model to make the representation dimension of the output result of the review feature extraction model and the audio feature extraction model consistent. The Mert model can refer to a large audio representation model that has already been trained.

[0053] Among them, the song audio features can refer to the features that characterize the audio content of the song to be scored. In practical applications, the song audio features can include data obtained by encoding the audio of the song to be scored.

[0054] In this context, a comment feature extraction model refers to a model used to determine the features of a comment. In practical applications, this model encodes the text information corresponding to the comment to obtain its features. The model can include a text representation model, which can then be combined with an MLP (Multilayer Perceptron) module or other neural networks used for data preprocessing, feature extraction, and representation learning to form the final comment feature extraction model. The text representation model can include, but is not limited to, models such as the BERT model (a language model) and word2vector (another language model). The MLP module can adjust the representation dimension of the text representation model's output to ensure consistency between the representation dimensions of the comment feature extraction model and the audio feature extraction model. It should be noted that if the target comment information is in audio format, it can be converted to text before the comment feature extraction model extracts the corresponding comment features.

[0055] Among them, comment features can refer to information that characterizes the content features of a comment.

[0056] As an example, the server can input the audio information of the first sample song in the sample song dataset into the audio feature extraction model to be trained. The audio feature extraction model can extract the audio features of the first sample song. Then, the server can input the comment information of the first sample song and the comment information of the second sample song into the comment feature extraction model to be trained. The comment feature extraction model to be trained can obtain the comment features of the first sample song based on the comment information of the first sample song, and the comment feature extraction model to be trained can obtain the comment features of the second sample song based on the comment information of the second sample song. In practical applications, the comment feature extraction model can include a BERT model and a first MLP module (a neural network). The BERT model can process the text information corresponding to the comment to obtain the mapping features corresponding to the comment. Then, the first MLP module processes the mapping features corresponding to the comment to obtain the comment features corresponding to the comment.

[0057] Step S206: Based on the similarity between the song's audio features and the comment features of the first sample song, and the similarity between the song's audio features and the comment features of the second sample song, train the song rating model to be trained, and obtain the trained song rating model.

[0058] As an example, the server trains a song rating model based on the similarity between the audio features of the first sample song and the comment features of the first sample song, as well as the similarity between the audio features of the first sample song and the comment features of the second sample song. The server then modifies the model parameters of the song rating model until the model loss function value corresponding to the song rating model is less than a preset loss function threshold, thus obtaining the trained song rating model.

[0059] In this embodiment, a song scoring model is trained using the audio information of sample songs and the corresponding comments. This allows the song scoring model to learn the correlation between song audio and corresponding comments. Based on the pre-trained song scoring model, the target comments corresponding to the songs to be scored are accurately selected. From the perspective of comments, the correlation between comments and song audio is analyzed to score the songs. This makes song scoring go beyond pure audio analysis and does not require a strict score quantification mechanism. By making full use of comment information, songs can be scored more flexibly and varied, thereby improving the accuracy of song scoring.

[0060] In some embodiments, the comment information of the first sample song includes a first type of comment information and a second type of comment information, and the comment information of the second sample song includes both the first type of comment information and the second type of comment information. The first type of comment information is obtained by the user through input from the music platform, and the second type of comment information is obtained based on expert experience. The training of the song rating model, based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, includes:

[0061] The song audio information of the first sample song, the first type of comment information of the first sample song, and the first type of comment information of the second sample song are input into the song scoring model to be trained. Based on the song audio information of the first sample song, the first type of comment information of the first sample song is obtained, and the first type of comment information of the second sample song is obtained. Similarly, the song audio information of the first sample song, the second type of comment information of the first sample song, and the second type of comment information of the second sample song are input into the song scoring model to be trained. Based on the song audio information of the first sample song, the second type of comment information of the first sample song is obtained, and the second type of comment information of the second sample song is obtained. The song scoring model to be trained is trained based on at least the song audio features, the first type of comment features, the second type of comment features, the first type of comment features, and the second type of comment features of the second sample song.

[0062] The first type can refer to information that represents the classification of the first comment information. In practical applications, the first type can represent the rating information that the user inputs for the song through the music platform.

[0063] The second type can refer to information that represents the classification of the second comment information. In practical applications, the second type can represent that the rating information is obtained based on expert experience (such as evaluation of the song by a musician).

[0064] The first type of comment feature refers to the feature obtained by the song rating model after processing the first type of comment information. The second type of comment feature refers to the feature obtained by the song rating model after processing the second type of comment information.

[0065] As an example, the server can train a song scoring model for both general evaluation and expert evaluation based on a large number of user comments (first type of comment information) on a music platform and second type of comment information obtained from expert evaluations of the sample songs. Specifically, the server can train the song scoring model simultaneously based on both first and second type of comment information. For example, the server inputs the audio information of the first sample song into the audio feature extraction model of the song scoring model to be trained to obtain the audio features of the first sample song, and inputs the first type of comment information corresponding to the first sample song and the first type of comment information corresponding to the second sample song into the comment feature extraction model of the song scoring model to be trained to obtain the first type of comment features of the first sample song and the first type of comment features of the second sample song. The server can train the song scoring model based on at least the audio features of the first sample song, the first type of comment features of the first sample song, and the first type of comment features of the second sample song to achieve training for general evaluation.

[0066] When training the song rating model for expert evaluation, the server can input the audio information of the first sample song into the audio feature extraction model of the song rating model to be trained, to obtain the audio features of the first sample song. It can also input the second type of comment information corresponding to the first sample song and the second type of comment information corresponding to the second sample song into the comment feature extraction model of the song rating model to be trained, to obtain the second type of comment features of the first sample song and the second type of comment features of the second sample song. The server can train the song rating model based on at least the audio features of the first sample song, the second type of comment features of the first sample song, and the second type of comment features of the second sample song, thus achieving expert evaluation training.

[0067] In this embodiment, the song rating model is trained based on sample audio and different types of comments. This allows the song rating model to learn the correlation between different types of comments and songs, optimize the performance of the song rating model, and thus improve the accuracy of the song rating output by the song rating model.

[0068] In an exemplary embodiment, a song rating model is trained based on at least the song audio features of a first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song, including:

[0069] The song rating model is trained based on the first similarity between the first sample song's audio features and the first type of comment features of the first sample song, the second similarity between the first sample song's audio features and the first type of comment features of the second sample song, the third similarity between the first sample song's audio features and the second type of comment features of the first sample song, and the fourth similarity between the first sample song's audio features and the second type of comment features of the second sample song.

[0070] As an example, the server can train the song rating model based on a large number of user comments (first type of comment information) on the music platform and second type of comment information obtained by evaluating the sample songs based on expert experience. Specifically, the server can first calculate the first degree of similarity between the audio features of the first sample song and the first type of comment features of the first sample song, the second degree of similarity between the audio features of the first sample song and the first type of comment features of the second sample song, the third degree of similarity between the audio features of the first sample song and the second type of comment features of the second sample song, and the fourth degree of similarity between the audio features of the first sample song and the second type of comment features of the second sample song. Then, the server can train the song rating model to be trained by combining the first, second, third, and fourth similarity.

[0071] In this embodiment, the song rating model to be trained is simultaneously trained based on the song audio features of the first sample song, the first type of comment features and the second type of comment features of the first sample song, as well as the first type of comment features and the second type of comment features of the second sample song. This enhances the generalization ability of the song rating model, enabling the accurate target comments to be selected using the song rating model. The song rating is then combined with the rating of the target comments to improve the accuracy of the song rating.

[0072] In some embodiments, a song rating model to be trained is trained based at least on the song audio features of a first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song, including:

[0073] Based on the first similarity between the audio features of the first sample song and the first type of comment features of the first sample song, and the second similarity between the audio features of the first sample song and the first type of comment features of the second sample song, the song scoring model to be trained is trained to obtain a preliminary trained song scoring model; based on the third similarity between the audio features of the first sample song and the second type of comment features of the first sample song, and the fourth similarity between the audio features of the first sample song and the second type of comment features of the second sample song, the preliminary trained song scoring model is trained.

[0074] As an example, the server can train the song scoring model separately for general evaluation and expert evaluation. For instance, the server can train the song scoring model in stages based on first-type and second-type comment information, respectively. For example, to further optimize the model's performance, the server can train the song scoring model using the first similarity between the audio features of the first sample song and the first-type comment features of the first sample song, and the second similarity between the audio features of the first sample song and the first-type comment features of the second sample song. This achieves the first stage of training (general evaluation training) to obtain a pre-trained song scoring model. Then, based on the pre-trained song scoring model obtained in the first stage, the server trains the pre-trained song scoring model and fine-tunes its parameters according to the third similarity between the audio features of the first sample song and the second-type comment features of the first sample song, and the fourth similarity between the audio features of the first sample song and the second-type comment features of the second sample song. This achieves the second stage of training (expert evaluation training), resulting in a fully trained song scoring model, thus enabling the song scoring model to achieve more professional song scoring.

[0075] In this embodiment, by training the song scoring model in stages, the model parameters of the song scoring model can be gradually adjusted. This allows the song scoring model to learn from expert evaluations in addition to general evaluations, thereby fine-tuning the model parameters, optimizing the performance of the song scoring model, and improving the accuracy of the song scores output by the song scoring model.

[0076] In one exemplary embodiment, a song scoring method is provided. This method is illustrated using an application to a server as an example. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0077] The song to be scored is input into a pre-trained song scoring model to extract the song's audio features; the pre-trained song scoring model is trained using the song scoring model training method described above.

[0078] The song to be scored refers to a song that needs to be scored. In practical applications, the song to be scored can include songs or audio recordings sung by users on music platforms (such as music platforms with karaoke functions). The song audio features refer to the features that characterize the audio content of the song to be scored. In practical applications, the song audio features can include data obtained by encoding the audio of the song to be scored.

[0079] As an example, when a user needs to rate a song performed by another user on a music platform, the user can upload the audio of the song to the server. The server receives the audio of the song and extracts the audio features of the song using an audio feature extraction model in a pre-trained song rating model.

[0080] In a pre-defined comment database, at least one comment whose comment features meet a pre-defined similarity condition with the audio features of the song to be scored is identified as the target comment; wherein, the comment features of the comment information in the comment database are obtained by a pre-trained song scoring model.

[0081] The comment database can refer to a pre-set database containing a number of comments. In practical applications, the comment information in the database can include comments based on expert experience, i.e., expert comments (such as text / audio comments made by experts on songs). Expert comments can be obtained by collecting comments on songs on music platforms based on expert experience. In practical applications, the content of expert comments can include appreciation and / or rating of the song. In specific implementation, the server can calculate the distance between the song's audio features and the comment features corresponding to each comment, and combine this with a preset distance threshold to select comments with distances less than the threshold as target comment information.

[0082] The target comment information can refer to comments that are related to the song to be scored. In practical applications, the comment features of the target comment information can meet a preset similarity condition with the audio features of the song to be scored. The target comment information can be in text or audio format.

[0083] As an example, a server can pre-collect expert reviews and build a review database. When a song needs to be rated, the server can obtain the review features of each review in the database and determine the similarity between the song's audio features and the review features of each review. The server then selects reviews whose similarity meets a preset condition as target reviews. In practical applications, the review database can contain review information and / or review features. The server can first obtain the review features of each review in the database and then determine the similarity between the song's audio features and the review features of each review. Alternatively, the server can directly determine the similarity between the song's audio features and the review features of each review based on the review features of each review.

[0084] Based on the rating information corresponding to each target comment, determine the song rating for the song to be rated.

[0085] In this context, rating information can refer to information that corresponds to the target comment and represents the rating value. In practical applications, each comment in the comment database can have a one-to-one corresponding rating information, and each rating information can represent a specific rating value.

[0086] Among them, song score can refer to information that represents the performance score of the song to be scored. In practical applications, song score can be used to judge the performance of the song to be scored. The higher the song score, the better the performance of the song.

[0087] As an example, the server can obtain the rating information corresponding to each target comment of the song to be rated. Then, based on the above rating information, the server calculates the song rating of the song to be rated. Specifically, the server can combine the weight and rating value of each target comment to calculate the song rating of the song to be rated.

[0088] In the aforementioned song scoring method, at least one target comment is identified from a pre-set comment database based on the audio characteristics of the song to be scored. Then, by comparing the comment characteristics of each comment in the database with the song's audio characteristics, target comments that meet pre-set similarity criteria to the song's audio characteristics are accurately selected from the database. Combined with the scoring information corresponding to each target comment, the song's score is determined. This method analyzes the correlation between comments and song audio from a commentary perspective, moving beyond pure audio analysis and eliminating the need for strict score quantification. By fully utilizing comment information, it allows for more flexible and varied song scoring, thereby improving the accuracy of song scoring.

[0089] In some embodiments, the song to be scored is input into a pre-trained song scoring model to extract the song audio features of the song to be scored, including: mapping the song audio information of the song to be scored to a preset space through the feature mapping network in the audio feature extraction model of the song scoring model to obtain the mapping features corresponding to the song audio information; and converting the mapping features corresponding to the song audio information into the song audio features of the song to be scored through the feature transformation network in the audio feature extraction model.

[0090] The audio feature extraction model refers to a model used to determine the audio features of a song to be scored. In practical applications, the audio feature extraction model can encode the audio information of the song to be scored to obtain the audio features of the song. The audio feature extraction model can include an audio representation model. After the audio representation model, an MLP (Multilayer Perceptron) module or other neural networks used for data preprocessing, feature extraction, and representation learning can be concatenated to form a review feature extraction model. The audio representation model can include, but is not limited to, the Mert model (a music analysis model) and the Hubert model (a speech analysis model). The MLP module can adjust the representation dimension of the output result of the audio representation model to make the representation dimension of the output result of the review feature extraction model and the audio feature extraction model consistent. The Mert model can refer to a large audio representation model that has already been trained.

[0091] In this context, the feature mapping network can refer to a model used to map the audio information of a song to a preset space. In practical applications, the feature mapping network can include any of the audio representation models such as the Mert model (a music analysis model) and the Hubert model (a speech analysis model). The preset space can refer to a vector space used to represent the audio features of the song.

[0092] Among them, feature transformation network can refer to a model that transforms the feature dimension of the mapping features corresponding to the audio information of a song. In practical applications, feature transformation network can include MLP module (a type of neural network).

[0093] As an example, the server can extract the audio features of a song using an audio feature extraction model within a song rating model. Specifically, the server can input the audio information of the song to be rated into a pre-trained audio feature extraction model. The feature mapping network in the audio feature extraction model can map the audio information of the song to be rated to a preset space, obtaining the mapped features corresponding to the song audio information. Then, the feature transformation network in the audio feature extraction model can convert the mapped features corresponding to the song audio information into the song audio features of the song to be rated. In a specific implementation, the feature dimension of the mapped features corresponding to the song audio information is different from the feature dimension of the comment features corresponding to each comment. The server can use a feature transformation network (such as an MLP module) to process the mapped features corresponding to the song audio information so that the feature dimension of the song audio features corresponding to the song audio information is the same as the feature dimension of the comment features corresponding to each comment.

[0094] In this embodiment, based on the feature extraction network and feature transformation network in the audio feature extraction model, the audio information of the song to be scored is gradually mapped to a preset space and converted into the audio features of the song to be scored. This enables the analysis of the song audio information to obtain accurate audio features, which facilitates the accurate calculation of the similarity between the song audio features and the comment features corresponding to each comment. Based on the similarity, accurate target comments are selected, thereby improving the accuracy of song scoring.

[0095] In some embodiments, determining the song score of the song to be scored based on the rating information corresponding to each target comment information includes: weighting and summing the rating information corresponding to each target comment information according to the weight information corresponding to each target comment information to obtain the song score of the song to be scored.

[0096] Among them, weight information can refer to information that characterizes the importance of each target comment.

[0097] As an example, each target comment can have corresponding rating information. The server can combine the weight information pre-set for each target comment to calculate the song rating of the song to be rated. For example, if the weight information of the first target comment A corresponding to the song to be rated is f1 and the rating information of the first target comment A corresponding to the song to be rated is x1, and the weight information of the first target comment B corresponding to the song to be rated is f2 and the rating information of the first target comment A corresponding to the song to be rated is x2, then the song rating R of the song to be rated can be expressed as R = f1*x1 + f2*x2. It can be understood that the sum of the weights corresponding to each target comment is 1. Therefore, when the weights corresponding to each target comment are equal, the song rating of the song to be rated can be expressed as the average of the rating information corresponding to each target comment.

[0098] In this embodiment, the score information of each target comment is weighted and summed by combining the weight information of each target comment. This allows for accurate calculation of the song score based on the weight of different comments, thereby improving the accuracy of song scoring.

[0099] In some embodiments, traditional song scoring schemes primarily use expert scoring to train the model. The model only receives the score as prior information; how that score is derived depends entirely on the model's autonomous learning. This learning process is extremely challenging, as expert scoring considers a comprehensive and in-depth range of perspectives, and the model may only grasp the surface level. Traditional methods place high demands on both the data and the model. Expert scores need to cover all score segments; otherwise, the model cannot learn the differences between segments. Furthermore, the rigor of expert scoring must be consistent; otherwise, a unified standard cannot be established. However, these two issues are difficult to simultaneously address in practice. To solve the problem of low accuracy in song scoring obtained using traditional methods… Figure 3 The diagram illustrates a flowchart of a song scoring method based on a cross-modal pre-trained large model. First, the server acquires training samples from the song scoring model during the first training phase. This first phase can primarily use a large amount of data to enable the song scoring model to learn a general and comprehensive scoring method. In practical applications, the evaluations from the first training phase can serve as general evaluations, which can be implemented based on sample songs and their corresponding comments within the training samples.

[0100] It is understandable that sample songs have corresponding comments on music platforms (such as user comments on the sample songs in the comment section). Although not all comments on music platforms are objective and accurate, they can roughly reflect the singing level of the sample songs. It is worth noting that there is a one-to-one correspondence between each sample song and its corresponding comments on the music platform (such as user comments). For example, each sample song and its corresponding comments on the music platform can form a data pair with a correlation.

[0101] In the first training phase, the server can input the audio information of each sample song and the corresponding first-type comment information (such as user comments) into the song rating model to be trained. The Mert model in the first branch of the song rating model processes the audio information of each sample song to obtain the corresponding mapping features (audio embedding). Then, the MLP module 1 in the first branch of the song rating model processes the corresponding mapping features (audio embedding) to obtain the audio features (audio embedding1) of each sample song. The BERT model in the second branch of the song rating model processes the corresponding first-type comment information of each sample song to obtain the corresponding mapping features (text embedding). Then, the MLP module 2 in the second branch of the song rating model processes the corresponding mapping features (text embedding) to obtain the first-type comment features (text embedding1) of each sample song. The feature dimensions of the mapping features output by the Mert model and the BERT model are inconsistent. This is addressed through the MLP... The module outputs song audio features and comment features with the same feature dimension. Then, the server calculates the contrast loss between audio embedding1 and text embedding2. The contrast loss function can be expressed as:

[0102] .

[0103] In this model, the numerator is the dot product of q and k+, representing the distance between vectors q and k+. The denominator is the sum of the dot products of all positive and negative examples. The parameter τ controls the range of the loss function. Here, q can be understood as the audio features of audio a corresponding to the sample song, and k+ represents the first type of comment information (such as user comments or general evaluations) corresponding to audio a. Audio a (its audio features) and k+ constitute a pair of positive examples, while a and the first type of comment information of other sample songs constitute multiple negative examples. Similarly, the first type of comment information of other sample songs and a also constitutes negative examples. The learning objective of the song rating model is to make the representation of audio embedding1 closer to its corresponding text embedding2 and farther from the text embedding2 of other audios, so that the song rating model can learn the correlation between sample songs and their comments (such as user comments or general evaluations).

[0104] The training of a song scoring model can also include a second training phase. After the song scoring model converges in the first training phase, it enters the second training phase. The second training phase can introduce expert evaluations. Expert evaluations are often more accurate than user comments, but there are fewer of them. Therefore, they are mainly used to fine-tune the model based on the first training phase, allowing the song scoring model to learn to score more professionally. Specifically, the server can input the audio information of each sample song and the corresponding second-type comment information (such as expert comments) into the song scoring model to be trained. The process of the song scoring model processing the audio information of each sample song and the corresponding second-type comment information (such as expert comments) can refer to the first training phase mentioned above, so that the song scoring model completes the training of the second training phase and obtains a pre-trained song scoring model. It can be understood that the server can also train the song scoring model without using two training phases, that is, the server can train the song scoring model using only general evaluations or expert comments.

[0105] After the song scoring model is trained, in order to obtain quantified scores, the server can set corresponding evaluation scores (such as rating information) for each expert comment in the preset comment database. The server can determine the comment features of each expert comment in the preset comment database based on the pre-trained song scoring model, and build the database by combining the evaluation scores corresponding to each expert comment in the preset comment database. When a song needs to be scored, the server can input the song's audio information and expert comments from a pre-set comment database into a pre-trained song scoring model. The first branch of the pre-trained song scoring model (e.g., the song feature extraction model) uses the Mert model and MLP module 1 to determine the song's audio features. The second branch of the pre-trained song scoring model (e.g., the comment feature extraction model) uses the BERT model and MLP module 2 to determine the comment features of each expert comment in the pre-set comment database. The pre-trained song scoring model calculates the distance between the song's audio features and the comment features of each expert comment, and determines the n comment features whose distance to the song's audio features is less than a preset distance threshold (or the n closest comment features to the song's audio features). Here, n can be a positive integer greater than or equal to 1, and n is a pre-set parameter that can be flexibly adjusted based on actual needs. The server calculates the song's score information based on the average of the evaluation scores of each expert comment corresponding to the above n comment features.

[0106] In this embodiment, the song scoring model is trained in two stages. This allows the model to learn general scoring and evaluation patterns from a large amount of non-expert data, and to learn more professional scoring methods from a small amount of expert scoring data. The overall architecture of the model is a dual-tower input structure: one input is the singing audio, and the other is the text input about the song performance. Both inputs are mapped to the same-dimensional embedding space after passing through the audio encoder and text encoder, respectively. Through contrastive learning, the network learns the correlation between the singing audio and the comments. In the prediction stage, the similarity between the audio to be predicted and each text in the comment database is calculated. The text with the highest similarity is taken as the evaluation result. After quantifying the evaluation result, the evaluation score is obtained. This fully utilizes the textual information of the comments, so that the song scoring is no longer limited to the pure audio level. Introducing textual information can more effectively guide the model to evaluate, and the score is more flexible and varied, eliminating the need for a strict score quantification mechanism to address the shortcomings of pure audio evaluation, thereby improving the accuracy of song scoring.

[0107] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0108] Based on the same inventive concept, this application also provides a song scoring model training apparatus for implementing the song scoring model training method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the song scoring model training apparatus provided below can be found in the limitations of the song scoring model training method described above, and will not be repeated here.

[0109] In one exemplary embodiment, such as Figure 4 As shown, a song rating model training device is provided. The song rating model includes an audio feature extraction model and a comment feature extraction model, comprising: an acquisition module 402, an extraction module 404, and a training module 406, wherein:

[0110] The acquisition module 402 is used to acquire the audio information of the first sample song and the comment information of the first sample song, and to acquire the comment information of the second sample song that is different from the first sample song.

[0111] The extraction module 404 is used to input the song audio information into the audio feature extraction model to be trained to obtain the song audio features of the first sample song, and to input the comment information of the first sample song and the comment information of the second sample song into the comment feature extraction model to be trained, so as to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song.

[0112] The training module 406 is used to train the song rating model to be trained based on the similarity between the song audio features of the first sample song and the comment features of the first sample song, and the similarity between the song audio features of the first sample song and the comment features of the second sample song, so as to obtain the trained song rating model.

[0113] In an exemplary embodiment, the comment information of the first sample song includes a first type of comment information and a second type of comment information; the comment information of the second sample song includes both the first type of comment information and the second type of comment information; the first type of comment information is obtained by the user through input from a music platform, and the second type of comment information is obtained based on expert experience; the training module 406 is further configured to input the song audio information of the first sample song, the first type of comment information of the first sample song, and the first type of comment information of the second sample song into a song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the first type of comment features of the first sample song based on the first type of comment information of the first sample song, and the first type of comment information of the second sample song. The first type of comment features of the second sample song are obtained; the song audio information of the first sample song, the second type of comment information of the first sample song, and the second type of comment information of the second sample song are input into the song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the second type of comment features of the first sample song based on the second type of comment information of the first sample song, and the second type of comment features of the second sample song based on the second type of comment information of the second sample song; the song rating model to be trained is trained based at least on the song audio features of the first sample song, the first type of comment features of the first sample song, the second type of comment features of the first sample song, the first type of comment features of the second sample song, and the second type of comment features of the second sample song.

[0114] In an exemplary embodiment, the training module 406 is further configured to train the song rating model to be trained based on a first degree of similarity between the song audio features of the first sample song and the first type of comment features of the first sample song, a second degree of similarity between the song audio features of the first sample song and the first type of comment features of the second sample song, a third degree of similarity between the song audio features of the first sample song and the second type of comment features of the first sample song, and a fourth degree of similarity between the song audio features of the first sample song and the second type of comment features of the second sample song.

[0115] In an exemplary embodiment, the training module 406 is further configured to train the song rating model to be trained based on a first similarity between the song audio features of the first sample song and the first type of comment features of the first sample song, and a second similarity between the song audio features of the first sample song and the first type of comment features of the second sample song, to obtain a pre-trained song rating model; and to train the pre-trained song rating model based on a third similarity between the song audio features of the first sample song and the second type of comment features of the first sample song, and a fourth similarity between the song audio features of the first sample song and the second type of comment features of the second sample song.

[0116] Each module in the aforementioned song scoring model training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0117] In one exemplary embodiment, a song scoring device is provided, comprising: a feature extraction module, a comment determination module, and a song scoring module, wherein:

[0118] The feature extraction module is used to input the song to be scored into a pre-trained song scoring model to extract the song audio features of the song to be scored; wherein the pre-trained song scoring model is trained by the song scoring model training method described above.

[0119] The comment determination module is used to identify at least one comment from a preset comment database that meets a preset similarity condition with the song audio features of the song to be scored, and use it as the target comment information; wherein, the comment features of the comment information in the comment database are obtained by the pre-trained song scoring model.

[0120] The song rating module is used to determine the song rating of the song to be rated based on the rating information corresponding to each of the target comment information.

[0121] In an exemplary embodiment, the feature extraction module is further configured to map the song audio information of the song to be scored to a preset space through the feature mapping network in the audio feature extraction model of the song scoring model, thereby obtaining the mapping features corresponding to the song audio information; and to convert the mapping features corresponding to the song audio information into the song audio features of the song to be scored through the feature transformation network in the audio feature extraction model.

[0122] In an exemplary embodiment, the song scoring module is further configured to perform a weighted summation of the scoring information corresponding to each of the target comment information based on the weight information corresponding to each of the target comment information, so as to obtain the song score of the song to be scored.

[0123] Each module in the aforementioned song scoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0124] This application also provides a song scoring model training apparatus for implementing the song scoring model training method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the song scoring model training apparatus provided below can be found in the limitations of the song scoring model training method described above, and will not be repeated here.

[0125] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a song scoring model training method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0126] Those skilled in the art will understand that Figure 5The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0128] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0129] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0131] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0132] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for training a song scoring model, characterized in that, The song rating model includes an audio feature extraction model and a comment feature extraction model, and the method includes: Obtain the audio information of a first sample song and its comment information, and obtain the comment information of a second sample song that is different from the first sample song; the comment information of the first sample song includes a first type of comment information and a second type of comment information; the comment information of the second sample song includes the first type of comment information and the second type of comment information. The song audio information is input into the audio feature extraction model to be trained to obtain the song audio features of the first sample song. The comment information of the first sample song and the comment information of the second sample song are input into the comment feature extraction model to be trained to obtain the comment features of the first sample song based on the comment information of the first sample song and the comment features of the second sample song based on the comment information of the second sample song. Based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, a song scoring model is trained to obtain a trained song scoring model. Specifically, during the training of the song scoring model, the model is trained based on the first similarity between the audio features of the first sample song and the first type of comment features of the first sample song, and the second similarity between the audio features of the first sample song and the first type of comment features of the second sample song, to achieve general evaluation training and obtain a pre-trained song scoring model. Then, based on the third similarity between the audio features of the first sample song and the second type of comment features of the first sample song, and the fourth similarity between the audio features of the first sample song and the second type of comment features of the second sample song, the pre-trained song scoring model is trained, and the model parameters are fine-tuned to achieve expert evaluation training, resulting in a trained song scoring model. Wherein, the first type of comment feature refers to the result obtained by the song rating model to be trained through processing the first type of comment information; the second type of comment feature refers to the result obtained by the song rating model to be trained through processing the second type of comment information; the first type of comment information is obtained by the user through input from the music platform, and the second type of comment information is obtained based on expert experience.

2. The method according to claim 1, characterized in that, The step of training a song rating model based on the similarity between the audio features of the first sample song and the comment features of the first sample song, and the similarity between the audio features of the first sample song and the comment features of the second sample song, to obtain the trained song rating model, includes: The song audio information of the first sample song, the first type of comment information of the first sample song, and the first type of comment information of the second sample song are input into the song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the first type of comment features of the first sample song based on the first type of comment information of the first sample song, and the first type of comment features of the second sample song based on the first type of comment information of the second sample song. The song audio information of the first sample song, the second type of comment information of the first sample song, and the second type of comment information of the second sample song are input into the song rating model to be trained, so as to obtain the song audio features of the first sample song based on the song audio information of the first sample song, the second type of comment features of the first sample song based on the second type of comment information of the first sample song, and the second type of comment features of the second sample song based on the second type of comment information of the second sample song. Based on the first similarity between the audio features of the first sample song and the first type of comment features of the first sample song, and the second similarity between the audio features of the first sample song and the first type of comment features of the second sample song, a song scoring model is trained to achieve general evaluation training, resulting in a pre-trained song scoring model. Then, based on the third similarity between the audio features of the first sample song and the second type of comment features of the first sample song, and the fourth similarity between the audio features of the first sample song and the second type of comment features of the second sample song, the pre-trained song scoring model is trained, and the model parameters are fine-tuned to achieve expert evaluation training, resulting in a fully trained song scoring model.

3. A song scoring method, characterized in that, The method includes: The song to be scored is input into a pre-trained song scoring model to extract the song's audio features; wherein the pre-trained song scoring model is trained by the method described in any one of claims 1 to 2. In a pre-defined comment database, each comment is identified as a target comment if its comment features meet a pre-defined similarity condition with the audio features of the song to be scored; wherein, the comment features of the comment information in the comment database are obtained by the pre-trained song scoring model. Based on the rating information corresponding to each of the target comment information, the song rating of the song to be rated is determined.

4. The song scoring method according to claim 3, characterized in that, The step of inputting the song to be scored into a pre-trained song scoring model to extract the song's audio features includes: The feature mapping network in the audio feature extraction model of the pre-trained song scoring model maps the song audio information of the song to be scored to a preset space to obtain the mapping features corresponding to the song audio information. The feature transformation network in the audio feature extraction model converts the mapping features corresponding to the song audio information into the song audio features of the song to be scored.

5. The song scoring method according to claim 3, characterized in that, When there are two or more target comment pieces, determining the song score of the song to be scored based on the rating information corresponding to each target comment piece includes: Based on the weight information corresponding to each of the target comment information, the score information corresponding to each of the target comment information is weighted and summed to obtain the song score of the song to be scored.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the song scoring model training method according to any one of claims 1 to 2 or the steps of the song scoring method according to any one of claims 3 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the song scoring model training method according to any one of claims 1 to 2 or the steps of the song scoring method according to any one of claims 3 to 5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the song scoring model training method according to any one of claims 1 to 2 or the steps of the song scoring method according to any one of claims 3 to 5.

Images