Multi-Stage Adaptive Audio Filtering for Noise Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-controlled devices face challenges in accurately performing automatic speech recognition due to background noise and limited capabilities in low-power devices with few microphones, especially in distinguishing and suppressing noise in real-time.
Innovation Solution
The implementation of an adaptive directionality system using multiple microphones, where one microphone is designated as primary for the target voice and another as reference for noise, with an adaptive filter that adjusts delay to enhance the target voice while suppressing ambient noise, allowing for more accurate speech recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple microphones are used with adaptive filtering to suppress noise, then speech recognition accuracy is improved, but device complexity increases
Solution Approach 1:
The noise suppression process is divided into multiple stages: first stage adaptive filtering to remove stationary noise, and second stage adaptive filtering to remove non-stationary noise. This segmentation allows complex noise suppression to be achieved through simpler, sequential processing steps, improving speech recognition accuracy while managing device complexity through structured decomposition of the filtering task.
Solution Approach 2:
The first stage adaptive filtering is applied preliminarily to remove stationary noise before the second stage processing. By performing preliminary noise suppression on stationary components first, the system prepares the signal for subsequent non-stationary noise removal, achieving cumulative noise suppression效果 while maintaining a manageable processing architecture.
2Measurement precision
If adaptive filtering is applied in real-time to suppress noise, then speech recognition accuracy is improved, but computational power requirements increase
Solution Approach 1:
The computational workload is segmented into two distinct filtering stages, each targeting specific noise characteristics. The first stage handles stationary noise with relatively simple adaptive filtering, while the second stage addresses non-stationary noise. This segmentation distributes computational requirements over time and allows for optimized processing at each stage, reducing peak power consumption while maintaining real-time performance.
Solution Approach 2:
The adaptive filtering operates in periodic cycles, alternating between first stage and second stage processing. This periodic action allows the system to process noise components in sequences rather than simultaneously, reducing instantaneous computational power requirements while achieving comprehensive noise suppression through repeated processing cycles.
Data Source
AI summary
The systems, devices, and processes described herein may include a first microphone that detects a target voice of a user within an environment and a second microphone that detects other noise within the environment. A target voice estimate and/or a noise estimate may be generated based at least in part on one or more adaptive filters. Based at least in part on the voice estimate and/or the noise estimate, an enhanced target voice and an enhanced interference, respectively, may be determined. One or more words that correspond to the target voice may be determined based at least in part on the enhanced target voice and/or the enhanced interference. In some instances, the one or more words may be determined by suppressing or canceling the detected noise.