Voice Activity Detection via Non-linear Noise Estimation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing natural language processing systems in mobile devices face increased power consumption and reduced battery life due to computationally expensive voice activity detection, which fails to accurately detect voice activity following sudden loud noises.

Innovation Solution

The system employs non-linearly updated background noise estimation for voice activity detection by calculating power spectral density distributions and using a Kullback-Leibler divergence-based update method to generate a reference PSD distribution, enhancing the accuracy of voice activity detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If voice activity detection is performed on the entire sound signal using traditional background noise estimation, then voice activity can be detected, but power consumption increases and battery life decreases

Engineering Contradiction:

Improvevoice activity detection accuracyVSAvoidpower consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent segments the sound signal into multiple frames and performs voice activity detection on each frame individually. This allows the system to process only relevant portions of the signal rather than the entire signal continuously, reducing computational load and power consumption while maintaining detection accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing voice activity detection only on frames where voice activity is likely present, rather than processing every frame uniformly. The system uses selective processing based on detected voice activity patterns to reduce overall computational expenses.

Inventive Principle:

Principle #16Partial or excessive action

2Loss of energy

If traditional background noise estimation is used, then computational costs are reduced, but voice activity following sudden loud noises cannot be detected

Engineering Contradiction:

Improvecomputational costVSAvoidvoice activity detection accuracy

Core Design Contradiction:

Loss of energyVSReliability

Solution Approach 1:

The patent implements dynamic background noise estimation that adapts to changing acoustic environments in real-time. The system continuously updates background noise models based on detected sound patterns, allowing it to respond to sudden loud noises and maintain accurate voice activity detection despite varying acoustic conditions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system uses feedback mechanisms where detected voice activity and sound patterns are fed back into the background noise estimation model. This continuous feedback loop allows the system to learn from recent acoustic events and improve its voice activity detection accuracy, particularly for sudden noise changes.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS11270720B2Background noise estimation and voice activity detection system

Publication Date: 2022.03.08 TEXAS INSTRUMENTS INC

AI summary

A method includes selecting a frame of an audio signal. The method further includes determining a first power spectral density (PSD) distribution of the frame. The method further includes generating a first reference PSD distribution indicating an estimate of background noise in the frame based on a non-linear weight, a second reference PSD distribution of a previous frame of the audio signal, and a second PSD distribution of the previous frame. The method further includes determining whether voice activity is detected in the frame based on the first PSD distribution of the frame and the first reference PSD distribution.