Audio Extraction Apparatus Center Cut Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio processing techniques fail to effectively separate accompaniment and vocal sounds from mixed musical audio signals, making it difficult to extract specific audio components for applications like karaoke or music education.
Innovation Solution
An audio processing system that uses a processor to perform preprocessing operations on stereo audio data, generating center cut audio data by determining the difference between two channels, and then employs a trained machine learning model or neural network to extract either the accompaniment or vocal sound by transforming the data into multi-dimensional spectrograms and separating the audio components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional audio processing techniques are used, then the processing is simple, but the separation accuracy of accompaniment and vocal sounds is poor
Solution Approach 1:
The patent segments the audio processing into distinct functional modules: preprocessing unit for channel separation and center cut audio generation, machine learning model for feature extraction and classification, and post-processing unit for audio reconstruction. This segmentation allows each module to specialize in specific tasks, improving overall separation accuracy while maintaining manageable system complexity through modular design.
Solution Approach 2:
The patent introduces center cut audio data as an intermediary element that captures the difference between stereo channels. This intermediary representation highlights the vocal portion while suppressing accompaniment, providing the machine learning model with enhanced features for accurate separation. The intermediary audio data serves as a bridge between the raw stereo input and the final separated outputs.
2Manufacturing precision
If no preprocessing is applied, then the processing is faster, but the extraction quality of specific audio components is poor
Solution Approach 1:
The patent applies preliminary preprocessing operations including channel separation and center cut audio generation before the main machine learning processing. By pre-processing the audio data to emphasize vocal features and suppress accompaniment, the system prepares optimized input for the machine learning model, improving extraction quality while reducing the computational burden during the main processing phase.
Solution Approach 2:
The preprocessing unit transforms the audio data by calculating the difference between stereo channels, effectively changing the parameter representation from raw stereo signals to center-cut audio features. This parameter transformation enhances the distinguishability of vocal and accompaniment components, enabling more accurate extraction by the machine learning model.
3Adaptability or versatility
If simple channel separation is used, then the device complexity is low, but the ability to isolate specific audio components is insufficient
Solution Approach 1:
The machine learning model is designed with multi-functionality to handle various audio separation tasks. The same model architecture can extract different audio components (vocal, accompaniment, specific instruments) by adjusting training data and parameters, providing universal applicability across different music genres and separation requirements without requiring separate specialized systems.
Solution Approach 2:
The system employs self-service mechanisms where the machine learning model automatically adapts to different audio inputs through its training data. By incorporating diverse training examples during model development, the system becomes self-sufficient in handling various audio separation scenarios without requiring manual configuration or complex rule-based systems for each case.
Data Source
AI summary
A processor in an audio extraction apparatus performs a preprocessing operation to determine, for a stereo audio source including first channel audio data including an accompaniment sound and a vocal sound for a first channel and second channel audio data including an accompaniment sound and a vocal sound for a second channel, a difference between the first channel audio data and the second channel audio data to generate center cut audio data, and an audio extraction operation to input the first channel audio data, the second channel audio data and the center cut audio data to a trained machine learning model to extract any one of the accompaniment sound and the vocal sound.