Multi-Encoder Music Processing for Selective Rhythm Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information processing systems using machine learning struggle to generate songs that match user-designated features, such as changing a melody while maintaining a certain rhythm, as they fail to selectively learn specific characteristics of the content.
Innovation Solution
An information processing apparatus employs multiple encoders to separately extract feature quantities from entire content and specific data, allowing for the generation of a learned model that can selectively learn user-designated features, such as rhythm, by using a first encoder for the entire content and a second encoder for specific data, and a decoder to reconstitute the content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single encoder is used to learn features of entire content, then the system can generate songs automatically, but it cannot selectively learn specific features designated by the user
Solution Approach 1:
The patent divides the feature learning process into two separate encoders: a first encoder that learns features from entire content and a second encoder that learns features from specific data. This segmentation allows the system to selectively learn specific features designated by the user while maintaining the overall song generation capability, directly resolving the contradiction between adaptability and complexity.
2Ease of operation
If machine learning learns all features of content, then comprehensive song generation is achieved, but user-designated feature control is lost
Solution Approach 1:
The patent extracts specific features from the entire content by using a second encoder that processes only specific data related to the user-designated feature. This extraction mechanism allows the system to maintain control over user-designated features while preserving the comprehensive information from the first encoder, thereby achieving both ease of operation and preventing information loss.
3Adaptability or versatility
If the system generates songs with all learned features, then complete song reproduction is achieved, but user-specified feature modification is impossible
Solution Approach 1:
The patent applies local quality by allowing different encoders to learn different aspects of the content with different levels of detail. The first encoder learns comprehensive features for complete song reproduction, while the second encoder focuses on specific user-designated features for accurate feature modification. This enables the system to achieve both adaptability in feature modification and precision in maintaining user-specified features.
Data Source
AI summary
An information processing apparatus according to the present disclosure includes an extraction unit that extracts first data from an element constituting first content, and a model generation unit that generates a learned model that has a first encoder that calculates a first feature quantity which is a feature quantity of first content, and a second encoder that calculates a second feature quantity which is a feature quantity of the extracted first data.