Voice Query Interpretation Using Dynamic Type Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice query systems struggle to accurately interpret and respond to dynamic types or non-static information in user queries, requiring users to rely on predetermined sets of clicks or typing exact information, limiting the ability to locate content through voice commands.
Innovation Solution
A system that utilizes Automatic Speech Recognition (ASR) to convert voice queries to text, employs natural language understanding (NLU) to detect dynamic types through segmentation and conditional random fields (CRF) analysis, generating tags based on query sequences and patterns, and uses user profiles and databases to enhance search accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system uses predetermined types and categories in the data graph, then the system structure is simple and stable, but the system cannot accurately interpret dynamic or non-static information in user queries
Solution Approach 1:
The system segments the query interpretation process into multiple stages: ASR converts voice to text, NLU detects dynamic types through segmentation and CRF analysis, and separate modules handle entity recognition and type detection. This segmentation allows the system to maintain a stable data graph structure while adding sophisticated dynamic type detection capabilities without overwhelming complexity.
Solution Approach 2:
The patent introduces an intermediary NLU layer between the ASR and the data graph search. This NLU layer acts as a mediator that detects dynamic types and generates appropriate search queries, bridging the gap between rigid predetermined types and flexible user queries without requiring fundamental changes to the underlying data graph structure.
2Ease of operation
If the system requires users to type exact information or use predetermined clicks, then the system maintains data accuracy, but the ease of operation deteriorates
Solution Approach 1:
The system replaces the mechanical interaction of typing and clicking with voice-based interaction. The ASR module converts spoken queries to text, and the NLU module interprets the intent, allowing users to search using natural speech rather than precise manual input, thereby improving ease of operation while maintaining interpretation accuracy through advanced NLP techniques.
Solution Approach 2:
The system enables self-service voice querying where users can independently search for dynamic content using natural speech without requiring knowledge of the system's internal type structure or predetermined categories. The NLU module automatically detects the intended meaning and formulates appropriate search queries.
3Adaptability or versatility
If the system uses static information storage in the data graph, then the system reliability is high, but the adaptability to dynamic queries deteriorates
Solution Approach 1:
The system introduces dynamic type detection capabilities that allow the NLU module to adapt to varying query types in real-time. The CRF-based detector dynamically identifies entity types and generates appropriate search queries based on the specific context of each user input, enabling the system to handle diverse dynamic queries while maintaining a stable underlying data graph.
Solution Approach 2:
The NLU module serves multiple functions: it performs entity recognition, detects dynamic types, generates search queries, and handles various query formats. This multi-functionality allows the system to maintain a reliable static data graph while adapting to diverse dynamic user queries through a single versatile interpretation layer.
Data Source
AI summary
The system receives a voice query at an audio interface and converts the voice query to text. The system identifies entities included in the query based on comparison to an information graph, as well as dynamic types based on the structure and format of the query. The system can determine dynamic types by analyzing parts of speech, articles, parts of speech combinations, parts of speech order, influential features, and comparisons of these aspects to references. The system combines tags associated with the identified entities and tags associated with the dynamic types to generate query interpretations. The system compares the interpretations to reference templates, and selects among the query interpretations using predetermined criteria. A search query is generated based on the selected interpretation. The system retrieves content or associated identifiers, updates metadata, updates reference information, or a combination thereof. Accordingly, the system responds to queries that include non-static types.