Network Microphone Wake-Word Detection With Two-Stage Keyword Spotting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Keyword spotting in network microphone devices is computationally demanding and power intensive, leading to increased power consumption and memory requirements, particularly in battery-powered devices, while also compromising accuracy and increasing the risk of false positives and negatives.
Innovation Solution
Implementing a two-stage keyword spotting algorithm, where a less demanding algorithm is used for preliminary detection, followed by a more accurate and resource-intensive algorithm to verify wake-word detection, and utilizing compressed neural network models for reduced memory and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single accurate keyword spotting algorithm is used continuously, then wake-word detection accuracy is improved, but power consumption and computational resource usage increase significantly
Solution Approach 1:
The keyword spotting process is divided into two distinct stages: a first stage using a computationally efficient algorithm for preliminary detection, and a second stage using a more accurate algorithm for verification. This segmentation allows the system to maintain high accuracy while reducing overall power consumption by avoiding continuous execution of the resource-intensive algorithm.
Solution Approach 2:
The system performs preliminary keyword spotting using a lightweight algorithm before engaging the more accurate but resource-intensive algorithm. This preliminary action filters out false positives early, ensuring that the heavier algorithm only processes promising candidates, thereby maintaining accuracy while conserving energy.
2Measurement precision
If a computationally intensive keyword spotting algorithm is used, then wake-word detection accuracy is improved, but device complexity and processing load increase
Solution Approach 1:
The processing pipeline is segmented into two algorithmic stages with different complexity levels. The first stage handles initial detection with simpler processing, while the second stage provides verification with more complex processing. This segmentation reduces overall device complexity requirements compared to continuously running a single high-complexity algorithm.
Solution Approach 2:
The system applies partial action by using the simpler algorithm for the majority of processing operations (initial detection of all audio streams), and only applies the complex algorithm partially when needed (verification stage). This approach reduces average processing complexity while maintaining high accuracy.
3Productivity
If continuous keyword spotting is performed to detect all wake words, then detection completeness is improved, but power consumption increases
Solution Approach 1:
The continuous detection process is segmented into alternating cycles of lightweight preliminary detection followed by selective heavy verification. This segmentation enables the system to maintain detection completeness across all audio streams while reducing average power consumption by keeping the heavy algorithm inactive during most time periods.
Solution Approach 2:
The system employs periodic action by cycling through the two algorithmic stages in a rhythmic pattern: preliminary detection runs continuously at low resource consumption, followed by periodic verification cycles of the accurate algorithm. This periodic execution maintains detection completeness while managing power consumption through temporal distribution of computational load.
Data Source
AI summary
Systems, methods, and devices with reduced power consumption in network microphone devices. In one embodiment, a network microphone device is configured to perform a method that includes (i) capturing audio content; (ii) using a first algorithm to perform a keyword detection process for determining whether the audio content includes a keyword; (iii) responsive to determining that the audio content includes the keyword, using a second, more computationally intensive algorithm to perform a wake-word detection process for determining whether the audio content includes a wake word; and (iv) responsive to performing the wake-word detection process, (a) causing a voice service corresponding to the wake word to process the audio content if the wake-word detection process confirms that the audio content includes the wake word or (b) ceasing performance of the wake-word detection process if the wake-word detection process disconfirms that the audio content includes the wake word.