Voice Recognition Selective Re-speak Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems often inaccurately transcribe spoken search queries, leading to irrelevant search results due to errors in recognizing user speech, which can be frustrating for users.
Innovation Solution
Implementing a dual speech recognition engine system where the first engine provides quick but less accurate processing for initial speech recognition, and a second engine, with higher accuracy, processes user corrections to the query, allowing users to re-speak incorrect portions for precise text generation and improved search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high-accuracy speech recognition engine is used, then recognition accuracy is improved, but processing time increases and system complexity increases
Solution Approach 1:
The speech recognition task is segmented into two parts: initial recognition using a fast engine, and correction of uncertain portions using a second engine. This divides the processing into stages, applying different levels of accuracy to different segments of the query based on confidence levels, thereby reducing overall processing time while maintaining accuracy where needed.
Solution Approach 2:
The system applies full accuracy processing only to portions of the speech query that are uncertain or incorrect, rather than processing the entire query with high accuracy. This partial application of the correction process reduces unnecessary computational overhead while maintaining accuracy for problematic segments.
2Measurement precision
If a single high-accuracy speech recognition engine is used, then recognition accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the speech recognition functionality into two distinct engines with different accuracy levels. The first engine handles initial recognition quickly, while the second engine handles corrections. This segmentation allows the system to achieve high accuracy without requiring all components to operate at maximum complexity continuously.
Solution Approach 2:
The system uses the output of the first speech recognition engine to identify which portions need correction, then automatically applies the second engine only to those portions. This self-service approach reduces the need for complex external control mechanisms to manage system complexity.
3Productivity
If speech recognition errors are not corrected, then processing speed is maintained, but search result relevance deteriorates
Solution Approach 1:
The system implements a feedback loop where the first speech recognition engine's output is evaluated, and uncertain or incorrect portions are identified and sent back for re-processing by the second engine. This feedback mechanism ensures that errors are corrected while maintaining overall processing efficiency by not re-processing entire queries unnecessarily.
Solution Approach 2:
Correction actions are applied partially only to the specific portions of the query that contain errors or uncertainty, rather than re-processing the entire query. This maintains processing speed by avoiding unnecessary re-processing while still improving search result relevance through targeted corrections.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Implementations of the present disclosure include actions of providing first text for display on a computing device of a user, the first text being provided from a first speech recognition engine based on first speech received from the computing device, and being displayed as a search query, receiving a speech correction indication from the computing device, the speech correction indication indicating a portion of the first text that is to be corrected, receiving second speech from the computing device, receiving second text from a second speech recognition engine based on the second speech, the second speech recognition engine being different from the first speech recognition engine, replacing the portion of the first text with the second text to provide a combined text, and providing the combined text for display on the computing device as a revised search query.