Speech Recognition Handling Missing Data Segments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition systems face accuracy issues due to missing speech segments, as current methods either misrecognize words or ignore missing syllables and phonemes, leading to user confusion and compounded errors.
Innovation Solution
A computer-implemented method that receives speech with missing segments, generates hypotheses for the missing segments, identifies the best hypothesis based on speech context, pronouncing lexicon, and language model, and inserts the best hypothesis to recognize or synthesize the speech, ensuring all hypotheses have an identical acoustic score.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If missing speech segments are ignored or extrapolated using traditional methods, then the speech recognition system can continue processing, but recognition accuracy deteriorates due to misrecognition of words with multiple missing segments
Solution Approach 1:
The patent segments the missing speech data into multiple candidate hypotheses rather than treating it as a single unknown entity. Each hypothesis represents a possible reconstruction of the missing segment, allowing the system to evaluate multiple possibilities simultaneously and select the most likely one based on acoustic scoring and language models.
Solution Approach 2:
The patent changes the parameter of acoustic scoring by assigning identical scores to all hypotheses for missing segments, effectively removing the acoustic penalty for missing data. This allows the language model and context to dominate the selection process, improving accuracy for words with missing segments while maintaining processing continuity.
2Measurement precision
If traditional speech recognition engines attempt to recognize words with missing segments, then some simple cases can be resolved, but complex cases with multiple missing segments lead to misrecognition
Solution Approach 1:
The patent introduces dynamic hypothesis generation and evaluation, where the system adapts its approach based on the specific pattern of missing segments. Rather than using a static recognition approach, the system dynamically creates and evaluates multiple hypotheses, selecting the best match based on combined acoustic and language model scoring.
Solution Approach 2:
The patent introduces an intermediary hypothesis evaluation stage between receiving incomplete speech and final recognition. This intermediary process generates multiple possible reconstructions of missing segments, scores them using both acoustic and language models, and selects the best hypothesis before final recognition, thereby improving reliability for complex cases.
3Ease of operation
If missing portions of speech are ignored as if they never existed, then processing can continue, but recognition results deteriorate due to missing syllables and phonemes
Solution Approach 1:
The patent performs preliminary action by generating multiple hypotheses for missing segments before the final recognition decision is made. This preliminary hypothesis generation and evaluation process ensures that all possible reconstructions are considered and scored, allowing the system to select the most accurate reconstruction before finalizing the recognition result.
Solution Approach 2:
The patent incorporates feedback mechanisms where the language model and acoustic scorer provide feedback on each hypothesis. The system uses this feedback to refine its selection, choosing the hypothesis that best fits both the acoustic patterns and language context, thereby improving accuracy while maintaining processing efficiency.
Data Source
AI summary
Disclosed herein are systems, computer-implemented methods, and tangible computer-readable media for handling missing speech data. The computer-implemented method includes receiving speech with a missing segment, generating a plurality of hypotheses for the missing segment, identifying a best hypothesis for the missing segment, and recognizing the received speech by inserting the identified best hypothesis for the missing segment. In another method embodiment, the final step is replaced with synthesizing the received speech by inserting the identified best hypothesis for the missing segment. In one aspect, the method further includes identifying a duration for the missing segment and generating the plurality of hypotheses of the identified duration for the missing segment. The step of identifying the best hypothesis for the missing segment can be based on speech context, a pronouncing lexicon, and/or a language model. Each hypothesis can have an identical acoustic score.