Decimated Bisectional Pitch Refinement for Audio Decoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for pitch estimation in speech and audio coding face complexity challenges, particularly in refining pitch estimates, which can lead to increased computational load and inaccuracies due to the need for precise time resolution and search range adjustments.
Innovation Solution
A bisectional search method combined with signal decimation is employed to refine pitch estimates, where the normalized correlation is evaluated at midpoints within a decreasing refinement range, reducing computational complexity while maintaining sufficient time resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If signal decimation is used to obtain a coarse pitch estimate, then the time resolution is reduced and computational complexity is lowered, but the pitch estimation accuracy deteriorates due to insufficient time resolution
Solution Approach 1:
The pitch estimation process is segmented into two distinct stages: a coarse pitch estimation stage using decimated signal to reduce computational complexity, and a refinement stage using the original undecimated signal to restore time resolution and improve accuracy. This segmentation allows each stage to optimize for its specific requirements.
Solution Approach 2:
The coarse pitch estimate obtained from the decimated signal serves as a preliminary result that guides the subsequent refinement process. By first obtaining a rough estimate, the system can then focus computational resources on refining only the critical portions of the pitch period, rather than performing full-resolution analysis throughout.
2Measurement precision
If the search range for pitch refinement is increased to account for pitch track deviation, then the accuracy of pitch estimation is improved, but the computational load increases
Solution Approach 1:
The refinement process applies different search ranges to different portions of the pitch period. A larger search range is used around the coarse pitch estimate to capture potential deviations, while the actual refinement computation is concentrated in the local region around the estimated pitch, rather than uniformly across the entire period. This localizes the computational effort where it is most needed.
3Productivity
If the decimation factor is increased to reduce computational complexity, then the processing speed is improved, but the time resolution of the pitch estimate deteriorates
Solution Approach 1:
The processing is segmented into two phases with different decimation factors: an initial coarse estimation phase using high decimation factor for speed, followed by a refinement phase using low or no decimation to restore time resolution. This allows the system to optimize for speed during coarse estimation and for accuracy during refinement.
Solution Approach 2:
The decimation factor is dynamically adjusted based on the processing stage and requirements. The system transitions from high decimation (fast processing) during coarse estimation to low or zero decimation (high time resolution) during refinement, making the decimation strategy adaptive rather than static.
Data Source
AI summary
A method and system for refining an estimated pitch period estimate based on a coarse pitch useful for performing frame loss concealment in an audio decoder as well as for other applications. A normalized correlation at the coarse pitch lag is computed and used as the current best candidate. The normalized correlation is then evaluated at the midpoint of the refinement pitch range on either side of the current best candidate. If the normalized correlation at either midpoint is greater than the current best lag, the midpoint with the maximum correlation is selected as the current best lag. After each iteration, the refinement range is decreased by a factor of two and centered on the current best lag. This bisectional search continues until the pitch has been refined to an acceptable tolerance or until the refinement range has been exhausted. During each step of the bisectional pitch refinement, the signal is decimated to reduce the complexity of computing the normalized correlation.