Far-Field Automatic Gain Control for Speech Recognition

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing automatic gain control methods struggle to effectively differentiate and adjust the gain of target and non-target signals in far-field speech signals, leading to suboptimal speech recognition accuracy.

Innovation Solution

An automatic gain control method that distinguishes between target and non-target signals in far-field speech signals by determining a gain table calculation parameter based on signal characteristics, such as energy ratios and double-talk judgment results, to adjust the gain accordingly.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If automatic gain control is applied to far-field speech signals, then speech recognition accuracy is improved, but the ability to differentiate between target and non-target signals deteriorates

Engineering Contradiction:

Improvespeech recognition accuracyVSAvoidsignal differentiation capability

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The patent segments the far-field speech signal into target signals and non-target signals through beamforming processing. The microphone array forms directional beams to separate signals from different spatial locations, enabling the system to identify and process target speech signals while suppressing non-target signals, thus resolving the contradiction between improving recognition accuracy and maintaining signal differentiation capability.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies different gain values to different signal components based on their spatial characteristics. By calculating gain values specifically for target signal directions while applying attenuation to non-target directions, the system achieves selective amplification that improves speech recognition accuracy without losing the ability to differentiate between target and non-target signals.

Inventive Principle:

Principle #3Local quality

2Reliability

If gain is increased for far-field speech signals, then signal quality is improved, but noise amplification increases

Engineering Contradiction:

Improvesignal qualityVSAvoidnoise amplification

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent applies different gain values to different spatial directions and signal components. By calculating gain values specifically for target signal directions and applying attenuation to non-target directions, the system achieves selective amplification that improves signal quality without amplifying noise, thus resolving the contradiction between improving reliability and reducing harmful noise amplification.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent converts the harmful effect of noise in far-field signals into a benefit by using spatial filtering and beamforming. The system identifies the spatial characteristics of noise and target signals, then applies gain control that suppresses noise while amplifying target speech, effectively converting the presence of noise into an opportunity for selective enhancement.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

3Loss of information

If beamforming is used to separate target and non-target signals, then signal separation is improved, but computational complexity increases

Engineering Contradiction:

Improvesignal separationVSAvoidcomputational complexity

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The patent segments the complex beamforming process into distinct stages: signal separation through spatial filtering, gain calculation for separated components, and selective application of gain values. This segmentation simplifies the computational process by handling target and non-target signals separately, reducing overall computational complexity while maintaining effective signal separation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies gain control selectively only to the separated target signal components rather than processing the entire mixed signal. By focusing computational resources on the specific target signal portions that need enhancement, the system achieves effective signal separation with reduced computational complexity compared to processing all signal components uniformly.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12283285B2Automatic gain control method and device, and readable storage medium

Publication Date: 2025.04.22 SOUNDAI TECH CO LTD

AI summary

An automatic gain control method and apparatus, and a readable storage medium. The automatic gain control method includes: for a far-field speech signal of a current frame, distinguishing between a target signal and a non-target signal; according to a result of the distinguishing between the target signal and the non-target signal, determining a gain table calculation parameter of the far-field speech signal, and obtaining a gain variation of the far-field speech signal of the current frame relative to a previous frame; determining a gain value for the far-field speech signal of the current frame according to the gain variation; and processing the far-field speech signal of the current frame according to the gain value determined, to obtain a processed speech signal. The automatic gain control method can effectively increase the gain of the target signal and reduce the gain of the non-target signal when gaining the far-field speech signal.