Pitch Lag Estimation Using Overlapping Autocorrelation Sections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pitch estimation methods face challenges in maintaining accurate and stable pitch tracks, especially in conversational speech coding, where complexity and delay requirements are stringent, and existing approaches often fail to provide optimal results.
Innovation Solution
The method involves determining autocorrelation values for an audio signal using two overlapping sets of sections within a delay range, allowing for improved accuracy and stability of pitch estimation by minimizing discontinuities and reusing computations, which can be implemented in hardware or software for various audio processing systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single set of sections is used for autocorrelation computations, then the complexity is reduced, but the pitch track stability deteriorates due to discontinuities at section boundaries
Solution Approach 1:
The delay range is divided into multiple sections, and autocorrelation computations are performed separately for each section. This segmentation allows for targeted processing that improves pitch track stability by ensuring continuous coverage across section boundaries while maintaining computational efficiency through localized computations.
Solution Approach 2:
Different sections of the delay range are processed with tailored autocorrelation computations appropriate for their specific characteristics. This local quality approach ensures that each section contributes optimally to the overall pitch estimation, improving stability without uniformly increasing complexity across the entire system.
2Measurement precision
If complex pitch estimation approaches are used, then the pitch estimation accuracy is improved, but the delay introduced for conversational applications increases
Solution Approach 1:
Autocorrelation computations are performed in advance for multiple sections within the delay range, preparing pitch lag candidates before they are needed for final pitch estimation. This preliminary action enables faster decision-making in conversational applications by having pre-computed values ready, reducing the critical processing delay while maintaining high accuracy.
Solution Approach 2:
The pitch estimation process is divided into separate autocorrelation computations for different sections, allowing parallel processing and optimization of each section independently. This segmentation reduces the overall processing time and delay by enabling efficient computation strategies for each section rather than processing the entire delay range as a single complex operation.
3Productivity
If the pitch lag search ranges are non-overlapping, then the computational efficiency is improved, but the accuracy of pitch estimation deteriorates due to discontinuities at boundaries
Solution Approach 1:
The problematic discontinuities at section boundaries are extracted and identified as the source of accuracy degradation. By recognizing these boundary effects, the patent applies targeted solutions such as overlapping sections or special boundary handling techniques that remove the harmful discontinuities while preserving the computational efficiency benefits of sectioned processing.
Solution Approach 2:
The patent introduces asymmetric overlapping between adjacent sections, where the search ranges extend beyond strict section boundaries. This asymmetric approach allows for continuous coverage and smooth transitions between sections, eliminating discontinuities and improving pitch estimation accuracy while maintaining computational efficiency through the structured asymmetric design.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Autocorrelation values are determined as a basis for an estimation of a pitch lag in a segment of an audio signal. A first considered delay range for the autocorrelation computations is divided into a first set of sections, and first autocorrelation values are determined for delays in a plurality of sections of this first set of sections. A second considered delay range for the autocorrelation computations is divided into a second set of sections such that sections of the first set and sections of the second set are overlapping. Second autocorrelation values are determined for delays in a plurality of sections of this second set of sections.