In-Vehicle Voice Recognition Using User-Specific DNN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In-vehicle intelligent terminals face low voice command recognition rates due to limited adaptability to different accents and user variations, leading to impediments in user utilization.
Innovation Solution
An in-vehicle voice command recognition method and apparatus utilizing a pre-trained deep neural network (DNN) model to acquire user basic information, recognize voice command contents, determine potential user intentions, and calculate confidence levels to execute corresponding actions, improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional voice recognition systems are used with limited command sets, then the system complexity remains low, but the recognition accuracy deteriorates due to inability to adapt to different accents and user variations
Solution Approach 1:
The system performs preliminary user identification and basic information extraction before voice command recognition. By pre-processing the input signal to extract user-specific characteristics and context information, the system prepares personalized recognition parameters in advance, thereby improving recognition accuracy for different accents and users without significantly increasing operational complexity
Solution Approach 2:
The system dynamically adjusts recognition parameters based on identified user characteristics and context information. By changing acoustic models, language models, and recognition thresholds according to user-specific parameters extracted in advance, the system adapts to different accents and speaking styles, thereby improving recognition accuracy while maintaining manageable system complexity through parameterized adaptation
2Measurement precision
If a pre-trained DNN model is introduced to determine user basic information and improve recognition accuracy, then the voice command recognition accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The DNN model is pre-trained offline with extensive user data and contextual information before deployment. By performing the computationally intensive training phase in advance, the system transfers learned patterns to the deployed model, enabling fast inference during actual voice command recognition without requiring real-time retraining, thus reducing processing time while maintaining high accuracy
Solution Approach 2:
The system extracts only the most critical user basic information and contextual features necessary for accurate recognition, rather than processing all possible attributes. By focusing on key discriminative features identified through the pre-trained model, the system achieves high recognition accuracy while minimizing processing time through selective feature extraction
3Reliability
If the system processes multiple potential user intentions with confidence level determination, then the recognition reliability improves, but the processing complexity increases
Solution Approach 1:
The system implements a feedback mechanism where confidence levels of multiple potential intentions are evaluated and compared. By using the confidence level information to iteratively refine intention selection and validate results against contextual constraints, the system improves recognition reliability through self-correction and verification, while managing complexity through structured feedback loops rather than exhaustive processing
Data Source
AI summary
An in-vehicle voice command recognition method and apparatus, and a storage medium. The method includes: acquiring a voice command inputted by a user; determining basic information of the user according to a pre-trained deep neural network (DNN) model; identifying contents of the voice command according to the basic information of the user, and determining at least one potential user intention according to the identified contents and a scenario page context at the time when the user inputs the voice command; determining a confidence level of the potential user intention according to the DNN model; determining a real user intention from the potential user intention according to the confidence level; and executing a corresponding action according to the real user intention. The embodiments of the present disclosure can effectively improve the correct recognition rate of voice commands.