Audio Coding via CNN Frequency Spectrum Recovery

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing audio coding methods, particularly at high bit-rates, face limitations in achieving high-quality sound reconstruction due to the loss of original frequency spectrum and inefficiencies in block-based frequency spectrum recovery, especially in the high-frequency band.

Innovation Solution

A method employing a neural network-based frequency spectrum recovery scheme that selectively transmits and recovers frequency spectral coefficients using a 2D check pattern, prioritizing the transmission of important signs and utilizing energy distribution information to improve coding efficiency and quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If parametric coding is used for low bit-rate coding, then coding efficiency is improved, but the ability to reconstruct high quality frequency spectral structure is lost

Engineering Contradiction:

Improvecoding efficiencyVSAvoidfrequency spectral structure reconstruction quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The frequency spectrum is segmented into multiple bands (low-frequency band and high-frequency band), with different coding strategies applied to each. The low-frequency band uses parametric coding for efficiency, while the high-frequency band uses transform coding for quality preservation, resolving the contradiction between coding efficiency and spectral reconstruction quality.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different coding methods are applied to different frequency regions based on their specific requirements. The low-frequency region receives parametric coding treatment while the high-frequency region receives transform coding treatment, allowing each region to be optimized locally for its specific needs rather than applying a uniform approach.

Inventive Principle:

Principle #3Local quality

2Reliability

If block-based frequency spectrum recovery is used in the high-frequency band, then recovery performance is improved, but the overall quality still remains lower than transform coding

Engineering Contradiction:

Improverecovery performanceVSAvoidsound quality

Core Design Contradiction:

ReliabilityVSManufacturing precision

Solution Approach 1:

Instead of trying to recover the high-frequency spectrum from low-frequency information (the conventional approach), the patent inverts the approach by directly encoding and transmitting the high-frequency spectral coefficients using transform coding, thereby achieving superior quality without relying on recovery algorithms.

Inventive Principle:

Principle #13The other way round (Inversion)

3Manufacturing precision

If all frequency spectral coefficients are transmitted, then sound quality is improved, but bit-rate increases

Engineering Contradiction:

Improvesound qualityVSAvoidbit-rate

Core Design Contradiction:

Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and transmits only the essential frequency spectral coefficients that contribute most to perceived sound quality, rather than transmitting all coefficients. By selecting and transmitting only the most important spectral information, the system achieves good sound quality at reduced bit-rates.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the representation parameters of the frequency spectrum by using parametric coding for certain bands and transform coding for others, allowing efficient compression while preserving quality. This parameter change enables adaptive bit-rate control based on quality requirements.

Inventive Principle:

Principle #35Parameter changes

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach enhances coding performance by reducing quantization errors and improving sound quality at high bit-rates, offering better results than traditional block-based recovery methods by leveraging data correlation in the 2D time-frequency domain.

Implementation Method 1

recovering a magnitude of a second frequency spectral coefficient through a neural network

Methodology Applied

Scientific EffectNeural network:

Data Source

PatentUS11508386B2Audio coding method based on spectral recovery scheme

Publication Date: 2022.11.22 ELECTRONICS & TELECOMM RES INST

AI summary

An inventive concept relates to an audio coding method to which CNN-based frequency spectrum recovery is applied. An inventive concept transmits a part of frequency spectral coefficients generated in transform coding to a decoder and the decoder recovers the frequency spectral coefficient not transmitted. Furthermore, the signs of frequency spectral coefficient are transmitted from an encoder to the decoder depending on a sign transmission rule.