Neural Sub-Band Audio Coding for Bitrate and Reconstruction Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio coding methods fail to efficiently control bitrate assignment and signal reconstruction for sub-band signals with varying perceptual relevance.
Innovation Solution
An audio signal encoding method that independently controls the reconstruction and bitrate allocation for a high band signal and a down-sampled core band signal using a neural network-based encoder and decoder system, comprising multiple encoding and decoding layers with specific stride values to maintain temporal resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sub-band signals are independently controlled for bitrate assignment and signal reconstruction, then coding quality is improved, but device complexity increases
Solution Approach 1:
The full-band signal is divided into multiple sub-bands (first sub-band signal and second sub-band signal) that are processed independently through separate encoding and decoding paths. This segmentation allows different bitrate allocations and reconstruction methods for each sub-band, improving overall coding quality while managing complexity through modular processing.
Solution Approach 2:
The patent introduces a hierarchical dimension to the encoding process by extracting feature vectors at different encoding layer depths (intermediate encoding layer for first sub-band, last encoding layer for second sub-band). This multi-dimensional feature extraction enables independent control of different signal components without proportionally increasing overall system complexity.
2Measurement precision
If feature vectors are extracted from intermediate and last encoding layers, then sub-band signal reconstruction is improved, but loss of information increases
Solution Approach 1:
The encoder neural network performs preliminary processing of the full-band signal through multiple encoding layers before feature vector extraction. By the time features are extracted from intermediate and last layers, the signal has already been transformed into a form that preserves essential information while reducing redundancy, minimizing information loss during subsequent compression.
Solution Approach 2:
The feature extraction process is nested within the encoding layers, where intermediate layer outputs feed into further processing that eventually produces the final encoded representation. This nested structure ensures that information is progressively refined and preserved through each processing stage, with later layers building upon the transformations of earlier layers.
Data Source
AI summary
An audio signal encoding/decoding method and an apparatus for performing the same are disclosed. The audio signal encoding method includes obtaining a full-band input signal, extracting a first feature vector corresponding to a first sub-band signal and a second feature vector corresponding to a second sub-band signal using an encoder neural network including a plurality of encoding layers, generating a first code vector corresponding to the first feature vector and a second code vector corresponding to the second feature vector by compressing the first feature vector and the second feature vector, and generating a bitstream by quantizing the first code vector and the second code vector.