Spectral Envelope Prediction for Audio Bandwidth Extension
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice or audio signal decoding methods, particularly in low-rate encoding, result in poor performance due to inadequate restoration of high-frequency bands using time-domain bandwidth extension, which affects the quality of the output signal.
Innovation Solution
A signal decoding method and device that predicts an excitation signal and spectral envelope for the extension band by selecting appropriate bands from the decoded signal, using spectral coefficients to determine a frequency-domain signal, thereby improving the performance of voice or audio signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If time-domain bandwidth extension is used to restore high-frequency bands, then encoding efficiency is improved, but the quality and performance of the output voice or audio signal deteriorates
Solution Approach 1:
The patent replaces the time-domain bandwidth extension method with a frequency-domain spectral envelope prediction method. Instead of using time-domain signal processing to restore high-frequency bands, the invention transforms the problem to the frequency domain where spectral envelopes are predicted and synthesized, achieving better signal quality while maintaining encoding efficiency.
Solution Approach 2:
The patent changes the domain of processing from time-domain to frequency-domain, and alters the parameters used for bandwidth extension. By predicting spectral envelopes in the frequency domain and using them to synthesize high-frequency bands, the method achieves superior signal quality compared to traditional time-domain approaches.
2Quantity of substance
If fewer bits are allocated to high-frequency bands during low-rate encoding, then encoding efficiency is improved, but the restoration quality of extended bands deteriorates
Solution Approach 1:
The patent introduces spectral envelopes as an intermediary element that bridges the gap between low-rate encoding and high-quality restoration. By predicting spectral envelopes from available coded bits and using them to guide the synthesis of high-frequency bands, the system achieves good restoration quality without requiring many coded bits for the high-frequency regions.
Solution Approach 2:
The patent performs preliminary prediction of spectral envelopes before synthesizing the extended high-frequency bands. This preliminary action of predicting the spectral characteristics allows the system to prepare the necessary information for high-quality restoration using minimal coded bits, resolving the contradiction between bit allocation and restoration quality.
Data Source
AI summary
A signal decoding method and device, where the method includes decoding a bit stream of a voice signal or an audio signal to acquire a decoded signal, predicting an excitation signal of an extension band according to the decoded signal, where the extension band is adjacent to a band of the decoded signal, and the band of the decoded signal is lower than the extension band; selecting a first band and a second band from the decoded signal, and predicting a spectral envelope of the extension band according to a spectral coefficient of the first band and a spectral coefficient of the second band; and determining a frequency-domain signal of the extension band according to the spectral envelope of the extension band and the excitation signal of the extension band.