Speech Recognition System Unknown Word Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In speech recognition systems, users face difficulties in identifying and correcting unrecognized 'unknown words' that do not exist in the recognition dictionary, as these words are falsely recognized and not displayed among lower candidates, leading to repeated incorrect recognition results.
Innovation Solution
The system includes a word recognition section, syllable recognition section, segment determination section, and output section that convert input speech to word and syllable sequences, determine segments corresponding to unknown words, and display nearby word entries in the dictionary, allowing users to easily identify and correct unrecognized words.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the system displays only first candidate and lower candidates of speech recognition results, then the display is simple and easy to understand, but unknown words that do not exist in the recognition dictionary cannot be identified or corrected
Solution Approach 1:
The patent extracts and highlights the specific segment corresponding to the unknown word from the recognition result, separating it from the rest of the text. This allows the system to maintain a simple display while explicitly indicating which part cannot be recognized, enabling users to identify and correct unknown words without cluttering the interface.
Solution Approach 2:
The patent uses visual highlighting (color changes) to mark the unknown word segment in the recognition result. This visual cue draws the user's attention to the problematic area without requiring additional text or complex interface elements, thus maintaining simplicity while providing necessary information about unrecognized words.
2Productivity
If the system repeatedly recognizes unknown words as existing dictionary words, then the recognition process continues smoothly, but the recognition accuracy deteriorates due to false recognition
Solution Approach 1:
The patent implements a feedback mechanism where the system highlights unknown word segments and provides users with the ability to correct them. This feedback loop allows users to intervene when false recognition occurs, improving recognition accuracy over time while maintaining continuous operation. The system learns from user corrections to improve future recognition of unknown words.
Data Source
AI summary
A speech recognition system in which, even when the user makes an utterance including a word that satisfies a predetermined condition such as an unknown word, such a fact can be presented to the user, and the user can confirm the fact easily, is provided. The speech recognition system includes a word speech recognition section that converts input speech to a recognition result word sequence by using a predetermined word dictionary for recognition, a syllable recognition section that converts input speech to a recognition result syllable sequence, a segment determination section that determines a segment that corresponds to a predetermined condition which is a ground for estimating that a word in the converted recognition result word sequence is an unknown word, and an output section that obtains a partial syllable sequence from the recognition result syllable sequence corresponding to the determined segment, and outputs one or more word entries, which are in the vicinity of a position at which the partial syllable sequence is arranged in the word dictionary for recognition in which words are arranged in the order defined for word entries, together with the recognition result word sequence.