Audio Bandwidth Extension Using GAM-Mapped High-Band Envelopes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current bandwidth extension (BWE) schemes for audio signals face challenges in achieving a balance between prediction accuracy and robustness, often introducing artifacts due to complex mappings, and struggle to effectively combine predicted and transmitted high band (HB) parameters.
Innovation Solution
A method using generalized additive modeling to map low band (LB) audio signal features to HB parameters, with a frequency shift and envelope control mechanism to estimate HB extensions, allowing for a balance between complex and constrained mapping schemes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex mapping schemes (e.g., based on GMM or HMM) are used to predict HB characteristics from LB features, then prediction accuracy is improved, but artifacts are introduced and robustness deteriorates
Solution Approach 1:
The patent changes the parameter representation from complex probabilistic models (GMM/HMM with many parameters) to a simpler linear regression model with fewer parameters. This parameter simplification reduces overfitting and improves robustness while maintaining prediction accuracy through careful selection of relevant LB features and optimization of the linear model parameters.
Solution Approach 2:
The patent segments the HB spectrum into multiple frequency bands and applies separate linear regression models to each band. This segmentation allows the system to capture local spectral characteristics more accurately while keeping each individual model simple, thereby improving overall prediction accuracy without introducing artifacts from overly complex global models.
2Measurement precision
If transmitted HB information is added to improve BWE performance, then quality is improved, but bit-budget consumption increases
Solution Approach 1:
Instead of transmitting the entire HB signal or all HB parameters, the patent transmits only selected critical HB parameters (such as spectral envelope parameters or excitation parameters) that are most important for reconstruction. This partial transmission approach improves quality by providing key HB information while consuming significantly less bit-budget than full HB transmission.
Solution Approach 2:
The patent uses LB features as an intermediary to indirectly represent HB characteristics. By extracting relevant features from the transmitted LB signal and using them to predict HB parameters through linear regression, the system avoids direct transmission of HB data while still achieving accurate HB reconstruction, thereby improving quality efficiency.
Data Source
AI summary
Estimation of a high band extension of a low band audio signal includes the following steps: extracting (S1) a set of features of the low band audio signal; mapping (S2) extracted features to at least one high band parameter with generalized additive modeling; frequency shifting (S3) a copy of the low band audio signal into the high band; controlling (S4) the envelope of the frequency shifted copy of the low band audio signal by said at least one high band parameter.


