Audio Encoder Matrix Factorization Reducing Quantization Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech coding technologies, such as ACELP, face significant computational complexity in quantizing the residual signal due to the need for explicit evaluation of all codebook entries, making them inefficient at higher bit-rates and suboptimal in terms of accuracy.
Innovation Solution
The proposed solution involves matrix factorization to decorrelate the objective function, using techniques like Vandermonde factorization, to reduce computational complexity and enable direct quantization of the residual signal, improving performance and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If brute-force search of the codebook is used to achieve optimal quantization, then signal-to-noise ratio is improved, but computational complexity increases to O(2^B N^2) which is unrealistic for B>=11
Solution Approach 1:
The patent segments the codebook search process by dividing the codebook into multiple subcodebooks or groups. Instead of exhaustively searching all 2^B entries, the algorithm performs a coarse search over a reduced set of representative vectors first, then refines the selection. This segmentation reduces the search space from O(2^B N^2) to a manageable complexity while maintaining acceptable quantization quality.
Solution Approach 2:
The patent applies partial action by performing quantization on a subset of codebook entries rather than all entries. The analysis-by-synthesis algorithm is applied selectively to promising candidates identified through preliminary criteria, such as energy distribution or spectral matching, rather than evaluating every possible codebook vector. This partial evaluation achieves near-optimal SNR with significantly reduced computational burden.
2Device complexity
If non-optimal quantizations are used to limit complexity, then computational complexity is reduced, but quantization accuracy deteriorates
Solution Approach 1:
The patent introduces an intermediary transformation step between the residual signal and the codebook quantization. A perceptual weighting filter or spectral transformation is applied as an intermediary that pre-processes the residual, making the subsequent quantization more effective. This intermediary operation aligns the signal characteristics with the codebook structure, improving quantization accuracy for the same computational complexity.
Solution Approach 2:
The patent changes parameters of the quantization process dynamically based on signal characteristics. The codebook search strategy, transformation type, or quantization step size is adjusted according to the local signal properties such as pitch period, spectral flatness, or energy level. These parameter changes enable the system to maintain high accuracy in difficult regions while using simpler methods in easier regions, balancing complexity and performance.
Data Source
AI summary
An encoder for encoding an audio signal has a predictor, a factorizer, a transformer and a quantize and encode stage. The predictor is configured to analyze the audio signal to obtain prediction coefficients describing a spectral analog of the audio signal or a fundamental frequency of the audio signal and subject the audio signal to an analysis filter function dependent on the prediction coefficients to output a residual signal of the audio signal. The factorizer is configured to apply a matrix factorization onto an audiocorrelation or covariance matrix of synthesis filter function defined by the prediction coefficients to obtain factorized matrices. The transformer is configured to transform the residual signal based on the factorized matrices to obtain a transformed residual signal. The quantize and decode stage is configured to quantize the transformed residual signal to obtain a quantized transformed residual signal or an encoded quantized transformed residual signal.