Neural Network Model for Accompaniment Purity Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for distinguishing between original and vocal cut accompaniments in music are inefficient and inaccurate, leading to a significant challenge in providing high-quality musical experiences.
Innovation Solution
A method utilizing a neural network model for accompaniment purity class evaluation, which involves extracting audio features from accompaniment data, training the model with labeled data, and adjusting model parameters to achieve high accuracy in distinguishing between pure instrumental and vocal cut accompaniments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual marking method is used to distinguish vocal cut accompaniment, then labor cost is high, but accuracy rate is low
Solution Approach 1:
The patent replaces the manual marking method (mechanical human operation) with an automated audio processing system that uses signal processing algorithms to extract features and classify accompaniment types. This substitution dramatically improves both efficiency (automated processing) and accuracy (consistent algorithmic classification)
Solution Approach 2:
The system enables accompaniment data to self-classify through automated feature extraction and classification algorithms, eliminating the need for manual human marking. The audio data itself provides the features needed for classification through signal processing, making the system self-sufficient
2Ease of manufacture
If vocal cut accompaniment is processed through audio technology, then original accompaniment can be obtained, but background noise increases
Solution Approach 1:
The patent applies preliminary classification to identify whether accompaniment data is original or vocal-cut before further processing. By detecting characteristic features of vocal-cut accompaniments (such as spectral patterns and temporal characteristics), the system can identify and filter out noisy processed accompaniments before they affect user experience
Solution Approach 2:
The system applies different processing and evaluation criteria to different types of accompaniment data. Original accompaniments are evaluated with higher purity standards while vocal-cut accompaniments are identified through their distinctive features and handled differently, allowing optimized processing for each type
3Measurement precision
If neural network model is trained with more data, then accuracy rate improves, but training time increases
Solution Approach 1:
The patent performs preliminary feature extraction and data preprocessing before model training, organizing audio data into standardized formats with extracted acoustic features. This preliminary preparation reduces the complexity of the training process and enables faster convergence even with large datasets
Solution Approach 2:
The system extracts key acoustic features from raw audio data before feeding them to the neural network model. By extracting only the most relevant features (such as spectral characteristics, temporal patterns, and statistical properties), the model trains faster while maintaining high accuracy
Data Source
AI summary
A method for accompaniment purity class evaluation and related devices are provided. Multiple first accompaniment data and a label corresponding to each of the multiple first accompaniment data are obtained, the label being used to indicate that corresponding first accompaniment data is pure instrumental accompaniment data or instrumental accompaniment data with background noise. An audio feature of each of the multiple first accompaniment data is extracted. Model training is performed according to the audio feature of each of the multiple first accompaniment data and the label corresponding to each of the multiple first accompaniment data, to obtain a neural network model for accompaniment purity class evaluation, a model parameter of the neural network model being determined according to an association relationship between the audio feature of each of the multiple first accompaniment data and the label corresponding to each of the multiple first accompaniment data.