Neural Audio Encoding with Recurrent and Nonrecurrent Mode Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based methods are ineffective in simultaneously removing both long-term and short-term redundancies in audio signals, with recurrent neural networks being inefficient for short-term redundancy and dimensionality reduction-based autoencoders failing to address long-term redundancy.
Innovation Solution
A method combining recurrent and nonrecurrent encoding/decoding models to effectively remove long-term and short-term redundancies by selectively applying different neural network-based models based on audio signal features, using a closed-loop or open-loop method to determine the encoding mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a recurrent neural network model is used to encode audio signals, then long-term redundancy is effectively removed, but short-term redundancy removal efficiency deteriorates
Solution Approach 1:
The patent segments the redundancy removal task into two distinct parts: a recurrent neural network model specifically for long-term redundancy removal, and a non-recurrent neural network model specifically for short-term redundancy removal. This segmentation allows each model to specialize in its respective strength, resolving the contradiction by preventing the recurrent model from being burdened with short-term redundancy tasks where it performs poorly.
Solution Approach 2:
The patent merges two different neural network models (recurrent and non-recurrent) into a unified encoding system that processes audio signals through both models. The recurrent model handles long-term temporal dependencies while the non-recurrent model handles short-term frame-level redundancies, and their outputs are combined to achieve comprehensive compression that overcomes the limitations of using either model alone.
2Loss of information
If a dimensionality reduction-based autoencoder is used to encode audio signals, then short-term redundancy is effectively removed, but long-term redundancy removal efficiency deteriorates
Solution Approach 1:
The patent segments the redundancy removal task into two distinct parts: a recurrent neural network model specifically for long-term redundancy removal, and a non-recurrent neural network model specifically for short-term redundancy removal. This segmentation allows each model to specialize in its respective strength, resolving the contradiction by preventing the recurrent model from being burdened with short-term redundancy tasks where it performs poorly.
Solution Approach 2:
The patent merges two different neural network models (recurrent and non-recurrent) into a unified encoding system that processes audio signals through both models. The recurrent model handles long-term temporal dependencies while the non-recurrent model handles short-term frame-level redundancies, and their outputs are combined to achieve comprehensive compression that overcomes the limitations of using either model alone.
3Device complexity
If only a single encoding model is used, then device complexity is reduced, but the ability to simultaneously remove both long-term and short-term redundancy deteriorates
Solution Approach 1:
The patent introduces dynamic mode selection that adapts the encoding process based on the characteristics of the input audio signal. The encoder dynamically determines whether to use recurrent mode, non-recurrent mode, or a combination of both modes depending on the signal properties, allowing the system to optimize performance for different types of audio content without requiring a permanently complex multi-model architecture.
Solution Approach 2:
The patent creates a universal encoding framework that can perform both long-term and short-term redundancy removal by integrating two different neural network models into a single versatile system. This multi-functional encoder can adaptively switch between or combine different encoding strategies, making it capable of handling diverse audio signal characteristics while maintaining a unified device structure.
Data Source
AI summary
The encoding method includes computing the first feature information of an input signal using a recurrent encoding model, quantizing the first feature information and producing the first feature bitstream, computing the first output signal from the quantized first feature information using a recurrent decoding model, computing the second feature information of the input signal using a nonrecurrent encoding model, quantizing the second feature information and producing the second feature bitstream, computing the second output signal from the quantized second feature information using a nonrecurrent decoding model, determining an encoding mode based on the input signal, the first and second output signals, and the first and second feature bitstreams, and outputting an overall bitstream by multiplexing an encoding mode bit and one of the first feature bitstream and the second feature bitstream depending on the encoding mode.