Phonetic Distance Measurement Using Speech Recognition Error Rates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for quantifying phonetic distance in speech recognition engines are limited by their reliance on physiological mechanisms and fail to accurately account for unique characteristics of specific speech recognition engines and speakers, particularly in handling insertion and deletion errors.
Innovation Solution
A system and method that compares recognized speech files with reference files to determine error occurrences and rates, calculating phonetic distances based on these errors, and normalizes them using a mapping function with coefficients to create a phonetic distance matrix that can be used for grammar selection and language training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If phonetic distance is estimated based on physiological mechanisms, then the measurement can be obtained using conventional methods, but the measurement fails to account for unique characteristics of specific speech recognition engines and speakers
Solution Approach 1:
The patent changes the measurement parameters from physiological-based estimates to actual error rates observed in speech recognition operations. By measuring substitution, insertion, and deletion errors specific to each engine-speaker combination, the system adapts phonetic distance measurements to actual performance characteristics rather than relying on conventional physiological models.
Solution Approach 2:
The system uses feedback from actual speech recognition error occurrences to refine phonetic distance measurements. By continuously monitoring substitution, insertion, and deletion errors in recognized speech files and comparing them against reference files, the system adjusts phonetic distance values to reflect real-world performance of specific engines and speakers.
2Reliability
If phonetic distance is measured using conventional physiological-based methods, then the process is simple, but it cannot handle insertion and deletion errors effectively
Solution Approach 1:
The patent segments the error analysis into three distinct categories: substitution errors, insertion errors, and deletion errors. Each error type is measured and weighted separately to calculate phonetic distance, allowing the system to handle different error mechanisms independently rather than using a single conventional metric.
Solution Approach 2:
The system introduces an intermediary comparison module that analyzes recognized speech files against reference files to identify and categorize errors. This intermediary process bridges the gap between raw speech recognition output and phonetic distance measurement, enabling reliable handling of insertion and deletion errors through systematic error detection and classification.
3Measurement precision
If phonetic distance measurements are normalized using mapping functions with coefficients, then the total separation between measured distances and existing matrices is minimized, but the normalization process adds computational complexity
Solution Approach 1:
The patent applies parameter changes through mapping functions that transform raw error rate measurements into normalized phonetic distance values. By using three normalization coefficients (a, b, c) in the mapping function d(i,j) = a + b/(e(i,j) - c), the system adjusts the scale and distribution of phonetic distances to minimize total separation from existing phonetic distance matrices while maintaining measurement accuracy.
Data Source
AI summary
Phonetic distances are empirically measured as a function of speech recognition engine recognition error rates. The error rates are determined by comparing a recognized speech file with a reference file. The phonetic distances can be normalized to earlier measurements. The phonetic distances/error rates can also be used to improve speech recognition engine grammar selection, as an aid in language training and evaluation, and in other applications.