Pitch Lag Selection for Stable Low-Complexity LTPF Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current low complexity pitch detection algorithms for Long Term Post Filtering (LTPF) in audio codecs are not satisfactory, particularly for complex signals like polyphonic music, resulting in unstable pitch contours due to jumps between local maxima of the weighted autocorrelation function.
Innovation Solution
A method involving two estimators to determine a pitch lag for an audio signal, where the first estimator finds the lag maximizing a correlation function for the current frame, and the second estimator searches within a subinterval containing the previous frame's pitch lag, with a selector choosing between the estimates based on normalized autocorrelation measurements to stabilize the pitch contour.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a low complexity pitch detection algorithm is used for LTPF encoding, then computational complexity is reduced, but pitch contour stability deteriorates due to jumps between local maxima
Solution Approach 1:
The pitch detection range is segmented into multiple subintervals, with each subinterval dedicated to detecting a specific harmonic component (fundamental frequency, second harmonic, third harmonic, etc.). This segmentation allows the algorithm to systematically search for local maxima in each subinterval and select the most appropriate pitch lag, thereby improving pitch contour stability while maintaining low computational complexity through targeted local searches rather than exhaustive global search.
2Measurement precision
If the pitch detection searches the entire lag range to find the global maximum, then measurement precision is improved, but computational complexity increases
Solution Approach 1:
Different subintervals are assigned to detect different harmonic components with specialized search strategies. The fundamental frequency subinterval searches for the primary pitch lag, while higher harmonic subintervals search for their respective local maxima. This local quality approach ensures that each subinterval's search is optimized for its specific harmonic component, achieving accurate pitch lag estimation for complex signals without requiring a computationally expensive exhaustive search of the entire lag range.
Solution Approach 2:
Instead of performing a complete exhaustive search of the entire lag range to guarantee finding the global maximum, the algorithm performs partial searches within multiple subintervals. By detecting local maxima in each subinterval and selecting the most appropriate one based on correlation measurements, the algorithm achieves sufficient accuracy for LTPF encoding without the excessive computational cost of a full global search.
3Stability of the object's composition
If a stable pitch contour is prioritized by using conventional algorithms, then pitch continuity is improved, but adaptability to complex signals deteriorates
Solution Approach 1:
The algorithm dynamically adapts its search strategy based on the signal characteristics in each frame. By dividing the lag range into multiple subintervals and detecting local maxima in each, the algorithm can flexibly respond to changing pitch conditions in complex signals like polyphonic music. The dynamic selection of pitch lag based on correlation measurements across subintervals allows the system to maintain stable pitch contours while adapting to varying signal content, resolving the contradiction between stability and adaptability.
Data Source
AI summary
In apparatus, methods, and programs for selecting pitch lag, an encoder obtains a first and a second estimates of a pitch lag for a current frame. A selected value is chosen by selection between the first and the second estimates, based on a first and a second correlation measurements. The second estimate is conditioned by the pitch lag selected at the previous frame. The selection is based on a comparison between: a downscaled version of a first correlation measurement associated to the current frame and obtained at a lag corresponding to the first estimate; and a second correlation measurement associated to the current frame and obtained at a lag corresponding to the second estimate.