Structured Repetition Model for Speech Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In voice-search applications, repeated spoken inputs are common due to speech recognition errors, leading to user frustration and inefficiency, with approximately half of interactions requiring repetition across various automated systems.
Innovation Solution
A structured model is used to analyze repeated utterances by performing joint probability analysis on word sequences and acoustic data, determining structural relationships such as exact matches, extensions, truncations, and spellings to improve speech recognition accuracy and identify user intentions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional speech recognition is used, then the system is simple and fast, but speech recognition accuracy is poor leading to frequent repetitions
Solution Approach 1:
The system performs preliminary analysis of the repeated utterance structure before final recognition. By detecting whether an utterance is a repetition and analyzing its structural relationship with the prior utterance (exact match, extension, truncation, or spelling variation), the system can apply appropriate correction logic in advance to improve recognition accuracy without requiring complex real-time processing during normal operation.
Solution Approach 2:
The system uses feedback from repeated utterances to correct recognition errors. When a user repeats an utterance, the system analyzes the structural relationship between the repeated and original utterances, using this feedback information to identify and correct recognition errors in the first utterance, thereby improving overall speech recognition accuracy.
2Measurement precision
If repeated requests are handled with simple repetition detection, then the system is easy to implement, but user intention is not accurately determined
Solution Approach 1:
The system segments the analysis of repeated utterances into distinct structural categories: exact matches, extensions (adding words before or after), truncations (removing words), and spellings (letter-by-letter repetition). By dividing the complex task of intention determination into these segmented analysis types, the system can accurately identify user intentions while maintaining manageable processing complexity through rule-based classification.
3Measurement precision
If joint probability analysis is performed on all word sequences and acoustic data, then speech recognition accuracy is maximized, but processing time and computational resources increase
Solution Approach 1:
The system applies local quality analysis by focusing joint probability computation on specific segments of the utterance that show structural relationships. Rather than analyzing all word sequences uniformly, the system identifies local regions where the repeated portion differs from the original (extensions, truncations, spellings) and applies intensive analysis only to these local areas, thereby maintaining high accuracy while reducing overall processing time.
Data Source
AI summary
Described is a technology by which a structured model of repetition is used to determine the words spoken by a user, and/or a corresponding database entry, based in part on a prior utterance. For a repeated utterance, a joint probability analysis is performed on (at least some of) the corresponding word sequences as recognized by one or more recognizers) and associated acoustic data. For example, a generative probabilistic model, or a maximum entropy model may be used in the analysis. The second utterance may be a repetition of the first utterance using the exact words, or another structural transformation thereof relative to the first utterance, such as an extension that adds one or more words, a truncation that removes one or more words, or a whole or partial spelling of one or more words.