Custom Wake-Up Word Detection for Smart Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wake-up systems for smart devices have a lower wake-up rate for custom wake-up words compared to main wake-up words, leading to poor user experience due to inadequate training data and potential false activations.
Innovation Solution
A method and apparatus that collect audio signals to detect custom wake-up words, determine a response strategy, and generate a response speech by either sending the custom wake-up word to a cloud server for a response text when connected, or using a preset or historical response text set when disconnected, allowing users to set custom wake-up words and responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If custom wake-up words are used, then user interaction is enhanced, but wake-up rate decreases compared to main wake-up words
Solution Approach 1:
The patent segments the wake-up recognition system into multiple independent recognition models, each trained for specific wake-up words (both main and custom). This allows the system to maintain high accuracy for custom wake-up words by dedicating specific computational resources to them, rather than relying on a single general model that may not perform well on custom words.
Solution Approach 2:
The system performs preliminary training of recognition models using pre-collected audio data containing custom wake-up words. By preparing and training models in advance with sufficient training data, the system ensures that custom wake-up words achieve high recognition accuracy before actual use, preventing the low wake-up rate problem.
2Adaptability or versatility
If custom wake-up words are detected, then user satisfaction increases, but false activations increase due to inadequate training data
Solution Approach 1:
The system performs preliminary training using pre-collected audio data that includes various scenarios and potential false activation triggers. By training in advance with comprehensive data, the recognition model learns to distinguish genuine custom wake-up words from similar sounds or phrases, significantly reducing false activations.
Solution Approach 2:
The system implements a feedback mechanism where recognition results are continuously monitored and used to improve the model. When false activations occur or when confidence scores are low, the system can request additional training data or adjust recognition thresholds, thereby reducing false positives over time while maintaining custom wake-up word detection capability.
Data Source
AI summary
Embodiments of the present disclosure relate to a method and apparatus for waking up a device. The method may include: collecting an audio signal in an environment the device located therein; determining, in response to that the audio signal includes a custom wake-up word being determined, a response strategy preset by a user and corresponding to the custom wake-up word; determining a target response text based on the response strategy; and generating a response speech of the target response text and playing the response speech.