Custom Hotword Evaluation for Accurate Voice Activation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech-enabled computing devices often use predetermined hotwords that may be inefficient or inaccurate due to acoustic confusability and user-specific pronunciation variations, leading to potential misactivation and resource wastage.
Innovation Solution
Users can define their own hotwords, and a system evaluates these using criteria such as word length, syllable count, edit distance, pronunciation variability, and training examples to generate a suitability score, providing feedback for selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If predetermined hotwords are used, then device activation is simplified, but acoustic confusability leads to detection inaccuracies and false positives
Solution Approach 1:
The system performs preliminary evaluation of candidate hotwords by analyzing acoustic confusability with other words in the vocabulary before finalizing the hotword selection. This advance assessment prevents selection of confusable hotwords, thereby improving detection accuracy while maintaining ease of operation through automated evaluation.
Solution Approach 2:
The system provides feedback to users about the suitability of candidate hotwords by indicating acoustic confusability issues. This feedback loop allows users to select alternative hotwords that are less confusable, thereby improving detection accuracy while maintaining simple operation through guided selection.
2Reliability
If user-defined hotwords are allowed, then hotword suitability improves for individual users, but evaluation complexity increases
Solution Approach 1:
Users define their own hotwords based on personal preferences or pronunciation patterns, and the system automatically evaluates these candidates using acoustic confusability analysis. This self-service approach improves reliability by allowing users to select familiar hotwords while the automated evaluation process manages complexity by objectively assessing suitability.
Solution Approach 2:
The system evaluates multiple parameters of candidate hotwords including acoustic confusability scores, pronunciation variability, and frequency of occurrence in speech corpora. By changing and analyzing these parameters, the system objectively assesses hotword suitability, improving reliability while managing evaluation complexity through systematic parameter analysis.
3Reliability
If hotword evaluation is performed, then false positives are reduced, but processing time increases
Solution Approach 1:
The system performs partial evaluation by focusing on the most critical aspect of hotword suitability - acoustic confusability with other vocabulary words. Rather than evaluating all possible characteristics, the system concentrates computational resources on the primary factor that causes false positives, thereby reducing evaluation time while maintaining effectiveness in preventing false activations.
Solution Approach 2:
The system pre-computes acoustic confusability metrics for candidate hotwords during the selection process, before actual hotword detection begins. This preliminary evaluation establishes baseline suitability scores that can be quickly referenced during operation, reducing the time required for real-time evaluation while maintaining false positive reduction capability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method and associated system, the method, when executed on data processing hardware, comprises causing the data processing hardware to perform operations comprising: receiving a user input indication indicating input of a candidate hotword by a user of a user device; determining an edit distance between a transcription of the candidate hotword and one or more other words that share a same language as the candidate hotword; generating, based on the edit distance, a hotword suitability score for the candidate hotword; determining whether the hotword suitability score for the candidate hotword satisfies a suitability threshold; and when the hotword suitability score for the candidate hotword satisfies the suitability threshold, accepting the candidate hotword as a hotword for the user device associated with the user and configuring the user device to enter a state to receive voice commands in response to utterances of the hotword.