Voice Trigger Noise Cancellation with Period Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional multi-microphone noise cancellers fail to effectively identify and discriminate between time segments with desired signals and noise, especially when noise is non-stationary and quasi-periodic, such as music or other voices, leading to delays in real-time applications like voice communication.

Innovation Solution

A method that detects voice triggers to identify noise and voice periods, updates noise and voice parameters based on audio signal features, and estimates the probability of voice presence to cancel noise in subsequent periods, using statistical estimation and buffering of relevant audio features.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional VAD algorithms based on stationarity and periodicity are used, then the system is simple to implement, but it fails when noise is non-stationary and quasi-periodic like music or other voices

Engineering Contradiction:

Improveability to handle non-stationary noiseVSAvoidcomplexity of noise discrimination algorithm

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary classification of time segments into noise-only, voice-only, and mixed segments before applying noise cancellation. This preliminary action allows the system to adapt to different noise conditions (including non-stationary noise like music) by preparing appropriate processing strategies in advance, resolving the contradiction between handling complex noise types and maintaining algorithm simplicity.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent divides the audio signal into distinct time segments and classifies each segment as noise-only, voice-only, or mixed. This segmentation allows the system to apply different processing strategies to different segments, enabling effective handling of non-stationary noise while keeping the overall algorithm structure simple and manageable.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If Blind-source-separation (BSS) methods are used to discriminate between desired voice and noise, then discrimination accuracy improves, but processing delay increases making it unacceptable for real-time applications

Engineering Contradiction:

Improveaccuracy of voice-noise discriminationVSAvoidprocessing delay

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent dynamically adjusts the noise cancellation strategy based on the classified segment type. For noise-only segments, full noise cancellation is applied; for voice-only segments, no cancellation is needed; for mixed segments, adaptive cancellation is applied. This dynamic adaptation enables real-time processing with accurate voice-noise discrimination without the delays associated with traditional BSS methods.

Inventive Principle:

Principle #15Dynamics

3Measurement precision

If the system buffers full audio signals for analysis, then accurate parameter estimation is achieved, but memory requirements and processing load increase significantly

Engineering Contradiction:

Improveaccuracy of parameter estimationVSAvoidmemory buffer requirements

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential features (spectral characteristics, temporal patterns, energy levels) from the audio signals during the preliminary classification phase, rather than buffering and processing the complete audio signals. This extraction approach maintains accurate parameter estimation for noise cancellation while significantly reducing memory buffer requirements and processing load.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS11120795B2Noise cancellation

Publication Date: 2021.09.14 DSP GROUP

AI summary

A method for noise cancellation, the method may include detecting a voice trigger or receiving an indication regarding an occurrence of a voice trigger; searching, based at least on a timing of occurrence of the voice trigger, for at least one of a noise-period and a voice-period; when finding a noise period then updating one or more noise parameters based on one or more features of sensed audio signals received during the noise period; when finding a voice period then updating one or more voice parameters based on one or more features of sensed audio signals received during the voice period; estimating, based on the one or more noise parameters and the one or more voice parameters, a probability that voice is present at one or more subsequent periods; and cancelling noise at the one or more subsequent periods, based on the voice presence probability.