Context-Aware ASR False Trigger Mitigation Using Location Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic speech recognition (ASR) systems suffer from false triggers due to ignoring additional data beyond audio input, leading to inefficient resource usage and privacy concerns, with existing mitigation methods being slow or incomplete.
Innovation Solution
A context-aware false trigger mitigation system that considers both audio embeddings and location data to determine the probability of a false trigger, using utterance length, differences in audio embeddings, and device location to rapidly identify non-intended interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If ASR systems process all audio inputs without additional context, then speech recognition coverage is maximized, but false triggers increase and resource usage becomes inefficient
Solution Approach 1:
The system segments the audio processing pipeline into two distinct stages: a lightweight audio embedding extraction stage that runs continuously, and a full ASR processing stage that activates only when false trigger probability is low. This segmentation allows the system to maintain high speech recognition coverage while reducing false triggers by filtering out unlikely candidates before full processing.
Solution Approach 2:
The system introduces an intermediary false trigger mitigation model that sits between audio input and full ASR processing. This intermediary layer uses audio embeddings combined with location data to calculate false trigger probability, acting as a filter that prevents unnecessary full ASR processing of likely false triggers while maintaining access to legitimate speech commands.
2Measurement precision
If ASR systems perform full processing on all audio inputs, then recognition accuracy is maintained, but resource consumption increases
Solution Approach 1:
The system applies partial action by performing only the necessary audio embedding extraction and false trigger probability calculation for all audio inputs, while reserving full ASR processing for only those inputs with low false trigger probability. This partial processing approach maintains recognition accuracy for legitimate commands while significantly reducing overall resource consumption by avoiding excessive full processing of false triggers.
Solution Approach 2:
The system performs preliminary action by extracting audio embeddings and calculating false trigger probability before initiating full ASR processing. This preliminary step uses computationally lighter operations to pre-filter audio inputs, ensuring that only promising candidates undergo resource-intensive full processing, thereby optimizing the balance between recognition accuracy and resource consumption.
3Device complexity
If ASR systems use only audio data for wake word detection, then system complexity is minimized, but false triggers cannot be effectively mitigated
Solution Approach 1:
The system merges multiple data sources by combining audio embeddings with location information to create a comprehensive false trigger probability assessment. This merging of audio and contextual data allows the system to maintain relatively simple audio processing while significantly enhancing false trigger mitigation capability through the addition of location-based filtering.
Solution Approach 2:
The system achieves multi-functionality by using the same audio embedding extraction infrastructure for both speech recognition and false trigger detection purposes. The audio embeddings serve dual functions: as features for wake word detection and as inputs for false trigger probability calculation, thereby enhancing false trigger mitigation without proportionally increasing system complexity.
4Speed
If ASR systems process audio continuously without location context, then responsiveness is maximized, but unnecessary processing occurs at incorrect locations
Solution Approach 1:
The system performs preliminary action by continuously extracting audio embeddings and calculating false trigger probability in advance, so that when a wake word is detected, the full ASR processing can be immediately activated without delay. This pre-computation of false trigger probability maintains responsive wake word detection while enabling rapid filtering of false triggers, thus preserving both response speed and useful processing throughput.
Solution Approach 2:
The system applies dynamics by making the ASR processing activation threshold dynamic rather than static. The false trigger probability calculation continuously adjusts the likelihood threshold based on location context, allowing the system to respond quickly to legitimate commands while dynamically filtering out false triggers. This dynamic thresholding maintains high response speed for valid inputs while reducing unnecessary processing at incorrect locations.
Data Source
AI summary
A method includes obtaining an audio input and a location associated with an electronic device. The method also includes generating an audio embedding associated with the audio input. The method further includes determining a first difference between the audio embedding associated with the audio input and an audio embedding associated with a known user. The method also includes determining a second difference between the location associated with the electronic device and a known location associated with the known user. The method further includes generating, using a false trigger mitigation (FTM) system, a probability of the audio input including a false trigger for automatic speech recognition based on the audio input, the first difference, and the second difference. In addition, the method includes determining whether to perform automatic speech recognition based on the probability.