Active Noise Cancellation With Keyword-Based Speech Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current active noise cancellation systems fail to distinguish between desired speech and unwanted noise, leading to poor user experience by either filtering out desired speech or amplifying unwanted noise, especially in environments like waiting rooms or airports.
Innovation Solution
A method and apparatus that generate a reference noise signal, identify keywords using speech recognition models, determine the direction of desired speech, extract the desired speech signal, and generate inverse noise to cancel external noise, utilizing a combination of Gaussian mixture models, Hidden Markov models, convolutional neural networks, beamformers, and principal component analysis to filter out unwanted noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If the existing ANC filter is used to cancel external noise, then the noise cancellation effect is improved, but the desired speech is incorrectly identified as noise and filtered out
Solution Approach 1:
The patent introduces an intermediary classification mechanism that processes the reference noise signal before ANC filtering. The system uses speech detection and keyword recognition as intermediary steps to classify noise signals into desired speech and unwanted noise categories. This intermediary classification layer prevents the ANC filter from incorrectly processing desired speech as noise, resolving the contradiction between noise cancellation and speech preservation.
Solution Approach 2:
The patent applies different processing qualities to different parts of the noise signal. By analyzing local characteristics such as speech patterns, keyword presence, and directional information, the system applies selective filtering - preserving segments containing desired speech while cancelling segments containing unwanted noise. This local quality differentiation allows simultaneous achievement of speech preservation and noise cancellation.
2Loss of information
If the full transparency mode is used to allow human voices to pass through, then the desired speech can be heard, but the unwanted noise cannot be cancelled
Solution Approach 1:
The patent implements local quality processing by analyzing specific characteristics of noise signals such as speech patterns, keyword presence, and directional information. Based on this local analysis, the system applies differentiated processing - allowing desired speech to pass through while cancelling unwanted noise in other segments. This resolves the contradiction by making the transparency mode selective rather than universal.
Solution Approach 2:
The system uses feedback mechanisms through adaptive filtering and continuous signal analysis. The classification results from speech detection and keyword recognition feed back into the filtering process, dynamically adjusting which signals are preserved and which are cancelled. This feedback loop enables the system to maintain desired speech while actively cancelling unwanted noise.
3Object-affected harmful factors
If the ANC filter processes all reference noise signals, then the noise cancellation coverage is improved, but the complexity of the system increases
Solution Approach 1:
The patent segments the noise processing task into distinct functional modules: speech detection module, keyword recognition module, directional determination module, and ANC filtering module. Each module handles a specific aspect of the processing, allowing the system to manage complexity through modular design while maintaining comprehensive noise cancellation coverage. The segmentation enables independent optimization of each module.
Data Source
AI summary
A method and an apparatus for noise reduction, an electronic device, and a storage medium are disclosed. The method includes: generating a reference noise signal by picking up external noise; identifying a keyword in the reference noise signal, and determining a direction of desired speech based on the identified keyword; extracting a desired speech signal in the direction of desired speech, and acquiring an unwanted noise signal by filtering out the desired speech signal from the reference noise signal; and generating inverse noise by filtering the unwanted noise signal, to cancel the external noise. Thus, this method not only makes it convenient for users to receive desired speech by effectively extracting the desired speech, but also improves the noise reduction effect by eliminating the unwanted noise.