Multi-Stage Hotword Detection for Low Power Voice Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In speech-enabled environments, existing systems face challenges in accurately distinguishing user-directed voice commands from background noise and unintended utterances without consuming excessive power or resources.
Innovation Solution
A multi-stage hotword detection system utilizing a digital signal processor for a coarse stage and software for a fine stage, where the coarse stage provides a quick approximation and the fine stage offers higher accuracy, allowing for efficient power usage and effective filtering of non-hotword utterances, with the fine stage being able to update via software.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single-stage high-accuracy hotword detection system is used, then detection accuracy is improved, but power consumption and processing time increase
Solution Approach 1:
The hotword detection system is divided into two independent stages: a coarse detection stage that performs quick initial filtering, and a fine detection stage that performs accurate verification only on promising candidates. This segmentation allows the system to achieve high overall accuracy while consuming less power, because the computationally intensive fine stage is activated only when the coarse stage identifies potential hotwords, not continuously on all audio input.
Solution Approach 2:
The coarse detection stage intentionally uses a lower accuracy threshold and more lenient matching criteria to quickly identify potential hotword candidates. This partial action approach accepts some false positives, but filters out the majority of non-hotword utterances early, preventing wasteful processing by the fine stage and thereby reducing overall power consumption.
2Reliability
If continuous processing of all utterances is performed, then detection reliability is improved, but processing time and computational resources increase
Solution Approach 1:
The coarse detection stage performs preliminary filtering of audio utterances before they reach the fine detection stage. By pre-processing and eliminating obvious non-hotword utterances early in the pipeline, the system prepares only relevant candidates for subsequent detailed analysis, thereby maintaining reliable detection while significantly reducing the time and computational resources required for full processing.
Solution Approach 2:
The coarse detection stage acts as an intermediary between the audio input and the fine detection stage. It mediates the processing flow by selecting which utterances deserve further attention, thereby optimizing the overall system performance and reducing unnecessary computational overhead on the fine stage.
3Device complexity
If a fixed detection algorithm is used, then system simplicity is maintained, but adaptability to different hotwords and environments decreases
Solution Approach 1:
The detection algorithm incorporates dynamic adaptability through configurable parameters and thresholds that can be adjusted based on different hotwords and environmental conditions. The coarse and fine detection stages use adaptive thresholding and parameter tuning, allowing the system to flexibly adapt to various hotwords and acoustic environments while maintaining a relatively simple overall architecture.
Solution Approach 2:
The system changes detection parameters such as sensitivity thresholds, matching criteria, and processing thresholds between the coarse and fine stages. These parameter changes enable the same basic detection architecture to handle different hotwords and environmental conditions effectively, providing versatility without requiring complete algorithm redesign.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for multi-stage hotword detection are disclosed. In one aspect, a method includes the actions of receiving, by a second stage hotword detector of a multi-stage hotword detection system that includes at least a first stage hotword detector and the second stage hotword detector, audio data that corresponds to an initial portion of an utterance. The actions further include determining a likelihood that the initial portion of the utterance includes a hotword. The actions further include determining that the likelihood that the initial portion of the utterance includes the hotword satisfies a threshold. The actions further include, in response to determining that the likelihood satisfies the threshold, transmitting a request for the first stage hotword detector to cease providing additional audio data that corresponds to one or more subsequent portions of the utterance.