Pitch Lag Selection for Stable Low-Complexity Audio Encoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Transform-based audio codecs introduce inter-harmonic noise, particularly at low delay and low bitrate, which affects the performance of harmonic audio signals, and existing pitch detection algorithms for Long Term Post Filtering (LTPF) are either too complex or produce unstable pitch contours, especially for complex signals like polyphonic music.

Innovation Solution

A method involving a first and second estimator to obtain pitch lag estimates for an audio signal, with the second estimator conditioned by the previous frame's pitch lag, and a selector to choose between these estimates based on normalized autocorrelation measurements, ensuring a stable and continuous pitch contour with reduced complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a pitch detection algorithm is performed at a low sampling rate to reduce complexity, then computational complexity is reduced, but pitch estimation accuracy and stability deteriorate

Engineering Contradiction:
Improvecomputational complexityVSAvoidpitch estimation accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The pitch detection process is segmented into two distinct estimation passes: a first estimate obtained by maximizing a first correlation function, and a second estimate obtained by maximizing a second correlation function in a neighborhood of the first estimate. This segmentation allows each estimator to focus on specific aspects of pitch detection, improving overall accuracy while maintaining low complexity through targeted searches rather than exhaustive analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The first estimate serves as a preliminary action that guides the second estimation process. By obtaining the first estimate first and then using it to define the search neighborhood for the second estimate, the algorithm performs preliminary work that simplifies subsequent processing. This preliminary action ensures that the final pitch selection benefits from both global correlation analysis and local refinement, achieving better accuracy without proportional increase in complexity.

Inventive Principle:
Principle #10Preliminary action

2Device complexity

If a single pitch estimate is selected without considering previous frames, then computational complexity is reduced, but pitch contour stability and continuity deteriorate

Engineering Contradiction:
Improvealgorithm complexityVSAvoidpitch contour stability
Core Design Contradiction:
Device complexityVSStability of the object's composition

Solution Approach 1:

The algorithm incorporates feedback from previous frame selections into the current frame's pitch detection. The second estimate is specifically obtained by maximizing the second correlation function in a neighborhood of the pitch lag selected for the previous frame. This feedback mechanism ensures that pitch estimates evolve smoothly across frames, maintaining contour stability and continuity while adding minimal computational overhead compared to independent frame-by-frame analysis.

Inventive Principle:
Principle #23Feedback

3Measurement precision

If exhaustive pitch lag search is performed to ensure accuracy, then pitch estimation precision is improved, but computational complexity increases

Engineering Contradiction:
Improvepitch lag estimation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The algorithm applies local quality by performing exhaustive correlation analysis only in a localized neighborhood around the first estimate, rather than across the entire possible pitch lag range. The second estimate maximizes the second correlation function specifically in this restricted neighborhood, which contains the pitch lag selected for the previous frame. This localized approach maintains high precision where it matters most while significantly reducing overall computational complexity.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The algorithm performs partial action by conducting two separate correlation maximizations instead of a single exhaustive search. The first correlation function provides a coarse estimate, and the second correlation function refines it locally. This partial approach to pitch detection—performing two targeted searches rather than one exhaustive search—achieves superior accuracy while keeping computational complexity manageable through strategic limitation of the search scope.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3707718B1Selecting pitch lag
Publication Date: 2021.10.13 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3707718B1 patent drawingFigure 1a
  • EP3707718B1 patent drawingFigure 1b
  • EP3707718B1 patent drawingFigure 2

AI summary

There are proposed techniques (e.g., in apparatus, methods, programs) for selecting pitch lag. An apparatus (10, 60a, 110) for encoding an information signal including a plurality of frames. The apparatus may comprise a first estimator (11) configured to obtain a first estimate (14, Ti), the first estimate being an estimate of a pitch lag for a current frame (13). The apparatus may comprise a second estimator (12) configured to obtain a second estimate (16, T2), the second estimate being another estimate of a pitch lag for the current frame (13). A selector (17) may be configured to choose (S103) a selected value (19, Tbest) by performing a selection between the first estimate (14, T1) and the second estimate (16, T2) on the basis of a first and a second correlation measurements (23, 25).