Speech Recognition Corpus Segmentation for Location Data Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in accurately recognizing location information, particularly in scenarios like map or navigation applications on mobile phones, resulting in low recognition rates and poor user experience due to the large amount of location data.
Innovation Solution
A method for speech recognition that extracts feature words associated with location information, calculates similarity with candidate words in a user-specific corpus containing standard and erroneous phonetically similar words, and outputs the correct standard candidate word based on preset correspondence when the similarity meets a predetermined condition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a traditional speech recognition system is used with a large corpus of location data, then the system can handle more location information, but the recognition accuracy decreases due to the large amount of data and phonetic similarities
Solution Approach 1:
The corpus is segmented into multiple sub-corpora based on user characteristics (e.g., regional accents, frequently visited locations). Instead of using a single large corpus, the system divides the location data into smaller, user-specific sub-corpora, making the recognition process more manageable and accurate for each user segment.
Solution Approach 2:
The system applies different recognition strategies and corpus selections to different users based on their local characteristics. Each user receives a customized recognition approach tailored to their specific needs, such as regional dialects or frequently visited locations, rather than a uniform approach for all users.
2Device complexity
If the corpus includes only standard location names, then the corpus size remains manageable, but the system fails to recognize phonetically similar erroneous speech inputs
Solution Approach 1:
The system pre-processes the corpus by identifying and adding phonetically similar erroneous location names alongside the standard location names. This preliminary action of expanding the corpus with potential error variants ensures that when speech recognition occurs, the system has already prepared the correct mappings for common pronunciation errors.
Solution Approach 2:
The system incorporates user correction feedback into the corpus. When users correct recognition errors, these corrections are used to update the corpus by adding the corrected location names and their phonetic variants, continuously improving the system's ability to recognize similar speech patterns in the future.
3Device complexity
If the system uses a generic recognition approach for all users, then the system complexity is low, but the user experience deteriorates due to incorrect recognition of user-specific speech patterns
Solution Approach 1:
The system dynamically adapts to each user's speech patterns and preferences. The corpus selection and recognition parameters are not static but change based on user behavior, allowing the system to evolve and improve its performance for each individual user over time without requiring complex manual configuration.
Solution Approach 2:
The system automatically improves itself by learning from user corrections and feedback. Users implicitly train the system by providing correction inputs, and the system uses this feedback to autonomously update its corpus and recognition parameters, eliminating the need for complex manual setup or ongoing user configuration.
Data Source
AI summary
A computer-implemented method for speech recognition is disclosed. The method includes extracting a feature word associated with location information from a speech to be recognized, and calculating a similarity between the feature word and respective ones of a plurality of candidate words in a corpus. The corpus includes a first sub-corpus associated with at least one user, and the plurality of candidate words include, in the first sub-corpus, a first standard candidate word and at least one first erroneous candidate word. The at least one first erroneous candidate word has a preset correspondence with the first standard candidate word. The method further includes in response to the similarity between the feature word and one or more of the at least one first erroneous candidate word satisfying a predetermined condition, outputting the first standard candidate word as a recognition result based on the preset correspondence.