Speech Query Engine for Structured Database Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional natural language processing techniques are inadequate for processing structured queries due to limitations in parsing and detecting semantics, especially in languages with unambiguous context-free grammar, leading to ambiguities and errors in speech recognition for devices like smartphones and IoT appliances.
Innovation Solution
A query processing engine that performs structure determination and literal determination on transcribed spoken queries using automatic speech recognition, identifying syntactically correct query strings by comparing tokens to ground truth query strings with minimal edit distance, and generating a rendered query string to correct errors and retrieve data from databases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional natural language processing techniques are used for speech recognition, then the system is simple and easy to implement, but the accuracy and reliability of structured query processing deteriorates due to ambiguities and errors
Solution Approach 1:
The query processing system is segmented into distinct modules: speech-to-text conversion module, structure determination module (with syntax validator and edit distance calculator), and literal determination module. This segmentation allows each module to specialize in specific tasks, improving overall reliability while managing complexity through modular design.
Solution Approach 2:
The patent introduces intermediate structures including syntax validation layer that checks transcribed queries against context-free grammar rules, and edit distance calculation that serves as an intermediary step to identify and correct transcription errors before final query execution.
2Ease of operation
If speech-based inputs are used instead of touch-based inputs, then ease of operation improves for lengthy or complex text strings, but measurement precision deteriorates due to transcription errors
Solution Approach 1:
The system implements feedback mechanisms where the transcribed query is validated against the syntax of the query language (e.g., SQL), and edit distance is calculated to identify potential transcription errors. This feedback loop allows the system to detect and correct inaccuracies in speech-to-text conversion, maintaining precision while preserving ease of operation.
Solution Approach 2:
The patent performs preliminary syntax validation and error correction on transcribed queries before execution. By checking the transcription against known query syntax patterns and correcting errors in advance, the system ensures high precision in the final query while maintaining the convenience of speech-based input.
3Productivity
If automatic speech recognition is used to transcribe queries, then productivity improves by enabling hands-free operation, but measurement precision deteriorates due to errors in transcribed queries
Solution Approach 1:
The patent replaces manual query writing with automated speech recognition, achieving hands-free operation and improved productivity. To address transcription accuracy, the system substitutes mechanical error correction with algorithmic approaches including syntax validation against context-free grammar and edit distance calculation to automatically identify and correct transcription errors.
Data Source
AI summary
A method may include receiving, from a client, a query dictated by a user. The query may request to retrieve, from a database, data stored in the database. A structure determination may be performed by identifying one of a plurality of ground truth query strings having a minimum edit distance relative to a transcription of the dictated query. Each the plurality of ground truth query strings may be consistent with at least a syntax of a query language associated with the dictated query. A rendered query string is generated by performing a literal determination that includes by comparing a literal token in the identified ground truth query string to one or more literal tokens present in the database. The rendered query string may be executed to retrieve, from the database, data requested by the dictated query string. Related methods and computer program products are also provided.