Embedded Noise Processing in Speech Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Voice-based wakeup systems for devices face challenges in power consumption and noise interference, particularly in distinguishing wake-up audio from background noise, which can lead to inefficient battery usage and degraded performance.
Innovation Solution
The implementation of a device with embedded noise processing in feature extraction using a Fourier Transform output, allowing for efficient noise filtering and speech recognition, reducing processing overhead and power consumption by integrating noise estimation and filtering with feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the device continuously listens for wakeup phrases in sleep mode, then wake-up detection capability is maintained, but power consumption increases
Solution Approach 1:
The audio processing pipeline is segmented into distinct stages: noise estimation during silent periods, speech detection when audio energy exceeds threshold, and feature extraction only when speech is detected. This segmentation allows the system to maintain wake-up detection capability while minimizing processing and power consumption during prolonged silent periods.
Solution Approach 2:
The system performs noise estimation periodically during silent intervals rather than continuously processing audio. Speech detection is triggered periodically based on energy threshold crossings, and full feature extraction is performed only periodically when speech is actually detected. This periodic action pattern reduces average power consumption while maintaining detection reliability.
2Object-affected harmful factors
If noise filtering is performed separately from feature extraction, then noise reduction effectiveness is improved, but processing complexity and power consumption increase
Solution Approach 1:
The patent merges noise filtering and feature extraction into a unified processing pipeline. The same audio buffer and FFT computations are reused by both the noise estimation module and the feature extraction module. Noise filtering parameters are computed and applied within the same processing context as feature extraction, eliminating redundant computations and reducing processing overhead while maintaining noise reduction effectiveness.
Solution Approach 2:
The audio processing infrastructure is designed to serve multiple functions: the FFT output is simultaneously used for noise spectrum estimation, speech feature extraction, and wake-up phrase detection. The same processing resources handle multiple tasks, reducing overall system complexity and power consumption while maintaining effective noise filtering capability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach effectively reduces power consumption and improves device wake-up performance by minimizing noise interference, enabling longer battery life and more accurate voice-based activation in noisy environments.
Implementation Method 1
both the noise filtering and the feature extraction use an output of the same Fourier Transform
Data Source
AI summary
Methods and an apparatus for performing feature extraction on speech in a microphone signal with embedded noise processing to reduce the amount of processing are provided. In embodiments, feature extraction and the noise estimate use an output of the same Fourier Transform, such that the noise filtering of the speech is embedded with the feature extraction of the speech.