Speech Coding LTP Codebook Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech encoding techniques, such as those using Linear Predictive Coding (LPC) and Code-Excited Linear Predictive (CELP) coding, face challenges in efficiently quantizing long-term prediction (LTP) coefficients, leading to high bitrate requirements and computational complexity due to the need for large codebooks and frequent updates.

Innovation Solution

The method employs multiple vector quantization codebooks to jointly quantize LTP coefficients, selecting the most efficient codebook based on rate-distortion trade-offs and using Entropy Constrained Vector Quantization, which reduces bitrate by using a single codebook indicator for all LTP vectors in a frame.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large codebooks are used for quantizing LTP coefficients, then quantization performance is improved, but bitrate and computational complexity increase

Engineering Contradiction:
Improvequantization performanceVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the codebook into multiple subcodebooks, each handling specific portions of the LTP coefficient space. This segmentation allows the system to achieve comprehensive quantization coverage without requiring a single large codebook, thereby reducing memory requirements and computational complexity while maintaining quantization performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic codebook selection where the appropriate subcodebook is chosen based on the current speech frame characteristics. This dynamic approach allows the system to adapt to varying speech conditions, using only the necessary portions of the codebook structure, thus reducing average computational complexity and bitrate while preserving quantization accuracy when needed.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If large codebooks are used for quantizing LTP coefficients, then quantization performance is improved, but bitrate increases

Engineering Contradiction:
Improvequantization performanceVSAvoidbitrate
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

By segmenting the codebook into multiple subcodebooks, the patent enables more efficient indexing and transmission. Instead of transmitting indices for a large monolithic codebook, the system transmits indices for smaller subcodebooks, reducing the overall bitrate required to achieve the same quantization performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by using only the necessary subcodebooks for each speech frame rather than the entire codebook structure. This selective approach transmits only the relevant quantization data, reducing bitrate while maintaining quantization performance for the actual speech content being encoded.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If codebooks are updated frequently, then encoding accuracy is improved, but computational complexity and storage needs increase

Engineering Contradiction:
Improveencoding accuracyVSAvoidstorage needs
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the codebook structure into multiple subcodebooks that can be independently managed and updated. This segmentation allows the system to maintain accurate encoding by updating only the necessary subcodebooks rather than the entire codebook, thereby reducing storage requirements and computational overhead while preserving encoding accuracy.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2384504B1Speech coding
Publication Date: 2018.07.04 SKYPE
  • EP2384504B1 patent drawingFigure 1a~1b
  • EP2384504B1 patent drawingFigure 2a~2b
  • EP2384504B1 patent drawingFigure 3~4

AI summary

A method, system and program for encoding and decoding speech according to a source-filter model whereby speech is modelled to comprise a source signal filtered by a time-varying filter. The method comprises: receiving a speech signal; and from the speech signal, deriving a spectral envelope signal representing the modelled filter and a remaining signal representing the modelled source. At intervals during the encoding, the method further comprises determining a period between portions of the remaining signal having a degree of repetition and determining a correlation between said portions based on that period, thus producing a respective vector of the correlation for each interval. Once every number of said intervals, the method further comprises selecting a codebook from a plurality of codebooks for quantizing the vectors, quantizing the vectors of that number of intervals according to the selected codebook, and transmitting the quantized vectors along with an indication of the selected codebook.