Electronic Device User Model Transfer and Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
New electronic devices often fail to properly understand user utterance intent, leading to incorrect operations, and require user training to comprehend voice inputs effectively.
Innovation Solution
An electronic device capable of receiving a pre-trained user model from another device, refining it, and identifying users based on voice inputs, with training also done using models from other users to enhance understanding and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a new electronic device is used without pre-trained models, then the device can operate immediately, but it fails to properly understand user utterance intent and requires extensive user training
Solution Approach 1:
The patent applies preliminary action by pre-training user models on other electronic devices before the user actually needs to use the new device. The pre-trained models containing user-specific speech patterns, vocabulary, and intent understanding are stored and can be quickly transferred to the new device, eliminating the need for extensive on-device training while maintaining high accuracy from day one.
Solution Approach 2:
The patent implements copying by replicating the pre-trained user models from one electronic device to another. The model data, which includes learned user characteristics and speech patterns, is copied across devices through communication networks, allowing the new device to inherit the user's interaction preferences and understanding without retraining.
2Measurement precision
If user models are trained on each electronic device individually, then the models are device-specific, but this requires extensive user training time and data collection on each new device
Solution Approach 1:
The system performs user model training in advance on the first electronic device, completing the time-consuming learning process before the user needs to use a new device. The pre-trained model captures user intent recognition accuracy and can be rapidly deployed to subsequent devices without requiring the user to spend time retraining on each new device.
Solution Approach 2:
The pre-trained user models are copied from the source device to the target device, transferring the accumulated training results. This copying mechanism preserves the high accuracy achieved through extensive training while eliminating the need to repeat the training process on each new device, thus saving significant time.
3Measurement precision
If pre-trained models are transferred from other devices, then user intent recognition accuracy improves immediately, but the device complexity increases due to model management and refinement processes
Solution Approach 1:
The patent introduces a server as an intermediary that manages the complexity of model transfer and refinement. The server handles model formatting, validation, and coordination between devices, reducing the computational and management burden on the electronic devices themselves while still enabling accurate pre-trained model transfer.
Solution Approach 2:
The system dynamically adjusts model parameters during the refinement process, optimizing the transferred model for the specific target device. By changing parameters such as model size, complexity, and configuration based on device capabilities and user needs, the system maintains high accuracy while adapting to different device constraints.
Data Source
AI summary
Disclosed is an operating method of an electronic device, including receiving a first user model from another electronic device of a user registered in the electronic device, through a communication circuit of the electronic device, refining a user model of the electronic device based on the first user model, and identifying the user based on a first voice input of the user by using the refined user model, wherein the user model is trained by the electronic device based on a second user model of at least one another user other than the user before the user is registered, and wherein the first user model is trained by the another electronic device based on a second voice input of the user obtained by the another electronic device of the user.