Audio Signal Resampling Using Predicted Future Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio frequency signal resampling methods, such as those used in the AMR-WB codec, introduce delays that can lead to audible artifacts and inefficiencies, particularly due to the use of zero padding to compensate for resampling delays, which are not effectively addressed in decoders and can result in periodic discontinuities.
Innovation Solution
The proposed method involves adaptive linear prediction to determine the number of future signal samples based on a chosen resampling delay, constructing a resampling support vector from current and predicted samples, and applying a resampling filter to reduce resampling delays while maintaining performance similar to continuous resampling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional FIR filter resampling is used to change sampling frequency, then filtering characteristics are improved, but resampling delay increases
Solution Approach 1:
The patent applies preliminary action by predicting future signal samples before they are actually available. The linear prediction module generates predicted future samples that are used in the resampling process ahead of time, allowing the resampling filter to operate with reduced delay while maintaining filtering quality. This is achieved by using the predicted samples as if they were actual future samples in the resampling calculation.
Solution Approach 2:
The patent introduces an intermediary element - the predicted future samples generated by linear prediction - that mediates between the current available samples and the future actual samples. These predicted samples serve as a bridge, allowing the resampling process to proceed with reduced delay while maintaining the filtering characteristics that would otherwise require access to actual future samples.
2Loss of time
If zero padding is used to compensate for resampling delay, then delay compensation is achieved, but periodic discontinuities and audible artifacts occur
Solution Approach 1:
The patent uses inexpensive, easily generated predicted samples as a temporary substitute for actual future samples. These predicted samples are computationally cheap to generate using linear prediction and are used only for the duration needed to reduce resampling delay. They are not permanent solutions but temporary placeholders that achieve delay reduction without the harmful effects of zero padding.
Solution Approach 2:
The patent changes the parameter values used in resampling by substituting actual future samples with predicted future samples. This parameter substitution allows the resampling process to operate with reduced delay while maintaining signal continuity. The predicted samples have different characteristics than zero values, thereby avoiding the periodic discontinuities and audible artifacts that result from zero padding.
3Measurement precision
If resampling is performed per frame with future signal prediction, then resampling accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent replaces the computationally intensive approach of actual future sample acquisition and storage with a simpler linear prediction mechanism. Instead of mechanically storing and processing actual future samples, the system uses mathematical prediction based on past samples, significantly reducing computational complexity while maintaining resampling accuracy.
Solution Approach 2:
The patent applies partial action by using linear prediction with a limited number of past samples to generate the needed future samples. Rather than using complex prediction models or excessive computational resources, the system uses a sufficient but not excessive amount of prediction based on recent signal history, achieving good accuracy with moderate computational effort.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for resampling an audio-frequency signal in an audio-frequency signal encoding or decoding operation. The method is such that it comprises the following steps per signal block to be resampled: - determining (E401), by adaptive linear prediction, a number of future signal samples, this number being defined on the basis of a chosen resampling delay; - constructing (E402) a resampling support vector from at least the samples of the current block and the determined future signal samples; -applying (E403) a resampling filter to the samples of the resampling support vector. The invention also relates to a resampling device implementing the described method, and an encoder and decoder comprising at least one device.