Adaptive VAD Threshold for Speech Processing in Noise

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech processing systems face challenges in accurately detecting voice activity in high noise environments due to fixed VAD thresholds, leading to high false positive and negative errors, as they struggle to adapt to varying noise levels and silence periods.

Innovation Solution

An adaptive voice activity detect (VAD) system that determines an average noise energy level during silence periods, converts it to sidetone attenuation, and generates an optimized VAD threshold based on the increase in voice level from sidetone attenuation, using a predetermined transfer function and time constants to adjust the threshold dynamically.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a fixed VAD threshold is used, then the system is simple to implement, but the VAD accuracy deteriorates in high noise environments due to high false positive and negative errors

Engineering Contradiction:

ImproveVAD threshold implementation complexityVSAvoidVAD accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent implements dynamic VAD threshold adjustment by continuously adapting the threshold based on measured noise levels. The system transitions from a fixed threshold to a dynamic threshold that changes with environmental conditions, specifically adjusting the threshold upward when noise levels increase to maintain accurate voice activity detection

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes the VAD threshold parameter based on noise level measurements. By monitoring the noise environment and adjusting the threshold parameter accordingly, the system maintains optimal detection accuracy across varying acoustic conditions without requiring complex fixed threshold tables

Inventive Principle:

Principle #35Parameter changes

2Reliability

If the VAD threshold is set low to detect speech in quiet environments, then false negative errors decrease, but false positive errors increase in high ambient noise situations

Engineering Contradiction:

Improvespeech detection reliability in quiet environmentsVSAvoidfalse positive errors in noise

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The system dynamically adjusts the VAD threshold based on real-time noise level measurements. In quiet environments, the threshold remains low to ensure speech detection, while in noisy environments, the threshold automatically increases to prevent false positives, thus adapting to environmental conditions

Inventive Principle:

Principle #15Dynamics

3Object-affected harmful factors

If the VAD threshold is set high to reduce false positives in noise, then false positive errors decrease, but false negative errors increase in quiet environments

Engineering Contradiction:

Improvefalse positive errors in noiseVSAvoidspeech detection reliability in quiet environments

Core Design Contradiction:

Object-affected harmful factorsVSReliability

Solution Approach 1:

The adaptive threshold mechanism dynamically adjusts the VAD threshold based on measured noise levels. When the environment is quiet, the threshold is set low to ensure speech detection; when noise increases, the threshold rises to reduce false positives, thus maintaining reliability across different acoustic conditions

Inventive Principle:

Principle #15Dynamics

4Adaptability or versatility

If the VAD threshold adapts in high noise environments, then the system can adjust to noise levels, but the threshold adaptation suffers errors and draws away from the optimal point due to low speech to noise ratio

Engineering Contradiction:

Improvethreshold adaptation capabilityVSAvoidthreshold adaptation accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system performs preliminary noise level measurement during silence periods before attempting threshold adaptation. By characterizing the noise environment in advance during non-speech periods, the system establishes a baseline that prevents erroneous adaptation when speech is present, thus maintaining adaptation accuracy in noisy environments

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS7881927B1Adaptive sidetone and adaptive voice activity detect (VAD) threshold for speech processing

Publication Date: 2011.02.01 HEWLETT PACKARD DEVELOPMENT COMPANY LP

AI summary

Systems and methods for adaptive sidetone and adaptive voice activity detect (VAD) threshold for speech processing are disclosed. The VAD system generally includes an adaptive VAD threshold generator configured to generate a VAD threshold based on an increase in voice level resulting from sidetone attenuation and a comparator for comparing received signals to the adaptive VAD threshold to determine the existence of voice activity. The sidetone attenuation is based on an average ambient noise energy level determined from a noise energy amplitude during periods of no voice activity and a comparator for comparing received signals to the adaptive VAD threshold to determine existence of voice activity.