Speech Encoding Correlation Sparsification for Lower Subframe Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing coding techniques for speech and sound signals in packet and mobile communications systems do not guarantee a reduction in worst-case computation amount per subframe without degrading the quality of the decoded signal during pitch period searches.
Innovation Solution
A coding apparatus and method that assesses the significance of each subframe and sets amplitude values of a predetermined number of samples to zero, adaptively adjusting the number of samples used for correlation calculations based on significance, ensuring reduced computation across all subframes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a threshold is adaptively varied from process to process to reduce computation amount, then the average computation amount across a frame is reduced, but the worst case computation amount per subframe is not guaranteed to be reduced
Solution Approach 1:
The patent applies preliminary action by computing the energy of each subframe before the pitch period search and using this pre-computed energy information to determine the number of samples to be processed in each subframe. This advance preparation allows the system to identify low-energy subframes beforehand and reduce computation in those specific subframes without affecting the overall quality, thereby guaranteeing reduced worst-case computation amount while maintaining signal quality.
2Productivity
If samples are sparsified by replacing values with zero when amplitude does not exceed a threshold, then computation amounts are reduced, but the quality of the decoded signal may be degraded
Solution Approach 1:
The patent applies local quality by making the sparsification process adaptive to local characteristics of each subframe. Instead of using a fixed threshold or uniform sparsification across all subframes, the system determines the number of samples to sparsify based on the energy of each individual subframe. This allows high-energy subframes to maintain more samples for better quality while low-energy subframes undergo more aggressive sparsification for reduced computation, thus maintaining overall signal quality while achieving computation reduction.
3Productivity
If a predetermined number of samples are set to zero in each subframe, then computation amount is reduced, but quality degradation occurs in low-energy subframes
Solution Approach 1:
The patent applies dynamics by making the number of samples to be processed dynamic rather than fixed. The system determines the number of samples to sparsify in each subframe based on the energy of that subframe, allowing the processing parameters to adapt dynamically to the characteristics of each subframe. This dynamic approach ensures that low-energy subframes do not suffer excessive quality degradation while still achieving computation reduction, as the sparsification level is adjusted according to actual signal energy rather than applying a fixed predetermined number.
Data Source
AI summary
An encoding device enables the amount of processing operations to be significantly reduced while minimizing deterioration in the quality of an output signal. This encoding device (101) encodes an input signal by determining the correlation between a first signal generated by using the input signal and a second signal generated by a predetermined method. An importance assessment unit (202) sets the importance of each of a plurality of processing units obtained by dividing the frames of the input signal. A CELP coder (203) performs sparse processing in which the amplitude value of a predetermined number of samples among multiple samples constituted by the first signal and/or the second signal in each processing unit is set to zero according to the importance that was set for each processing unit, and calculates the correlation between the first signal and the second signal, either of which was subjected to sparse processing.