Flexible Speech-to-Text Search Using Edit Distance Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech-to-text mechanisms struggle with transcribing uncertain speech due to variations in voices, accents, and out-of-vocabulary words, often requiring exact matches that lead to incorrect results when dealing with slang, professional terms, or mumbled words.
Innovation Solution
Implementing a 'fuzzy' search system that allows for approximate matches by calculating the edit distance between search terms and dictionary entries, selecting similar words with the smallest edit distances, and searching for these in text data to provide a range of close matches, including permutations and substitutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the STT dictionary is expanded to include out of vocabulary words, then the ability to transcribe uncertain speech is improved, but memory resources increase and search time becomes cumbersome
Solution Approach 1:
The patent segments the search process into two distinct phases: first searching the compact original dictionary for exact matches, and only if no match is found, then performing a secondary search using phoneme-based approximate matching. This segmentation allows the system to maintain a small dictionary while still handling out-of-vocabulary words efficiently, resolving the contradiction between transcription accuracy and dictionary size.
Solution Approach 2:
The patent introduces phoneme sequences as an intermediary layer between the original dictionary and the speech input. By converting both dictionary words and search terms into phoneme representations, the system enables approximate matching without expanding the dictionary, thus improving transcription reliability while maintaining manageable dictionary complexity.
2Speed
If the STT mechanism requires exact matches to dictionary entries, then search speed is maintained, but incorrect transcription occurs for words outside the dictionary
Solution Approach 1:
The patent performs a preliminary exact-match search in the original dictionary before attempting approximate phoneme-based matching. This preliminary action ensures that common words are transcribed quickly and accurately, while only uncertain or out-of-vocabulary words trigger the more computationally intensive phoneme-based search, thus maintaining overall search speed while improving transcription accuracy for difficult cases.
3Reliability
If the STT mechanism lowers matching standards to approximately match dictionary entries, then out of vocabulary words are captured, but incorrect search results are generated
Solution Approach 1:
The patent changes the matching parameter dynamically based on the search context. For exact-match searches, it uses strict string equality comparison. For out-of-vocabulary words, it transitions to phoneme-based approximate matching with configurable similarity thresholds. This parameter change allows the system to capture out-of-vocabulary words while maintaining match accuracy through controlled flexibility in the matching criteria.
Data Source
AI summary
A system and method for receiving an initial search entry to search text data. The text data may be, for example, an N word lattice, transcribed by a text to speech engine. The difference between the initial search entry and one or more entries in the dictionary may be measured. One or more similar entries may be selected from the dictionary that have the smallest measures of difference to the initial search entry. The text data may be searched for the one or more selected similar entries. Each of the searched similar entries found in the text data may be displayed as a search result.