Word Spotting Acoustic Event Comparison via Subword HMMs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing word spotting systems face challenges in accurately comparing acoustic events, leading to high false alarms and reduced detection rates due to imperfect similarity assessments between acoustic signals.
Innovation Solution
The method employs subword unit models, specifically Hidden Markov Models, to compute probability distributions over time, using algorithms like Forward, Backward, and Viterbi, and dynamic time warping to align and compare acoustic events, with Kullback-Leibler distance as a measure of dissimilarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional acoustic event comparison methods are used in word spotting systems, then the system complexity remains manageable, but the measurement precision of acoustic event similarity is insufficient leading to high false alarms and reduced detection rates
Solution Approach 1:
The patent segments acoustic events into discrete subword units and models each unit separately using Hidden Markov Models. This segmentation allows for more precise comparison by analyzing individual acoustic units rather than treating entire events as monolithic blocks, thereby improving measurement precision while managing complexity through modular modeling approaches.
Solution Approach 2:
The patent transforms acoustic event comparison from direct signal matching to probability distribution comparison using HMM parameters. By changing the parameter space from raw acoustic features to probabilistic state distributions, the system achieves higher measurement precision in similarity assessment while working within manageable computational frameworks.
2Reliability
If more sophisticated comparison methods are employed to improve detection accuracy, then the detection rate increases, but the computational time and processing complexity increase
Solution Approach 1:
The patent performs preliminary action by pre-computing probability distributions over subword units and storing HMM parameters before actual event comparison. This advance preparation enables faster real-time comparison operations, improving detection reliability without proportionally increasing processing time during query execution.
Solution Approach 2:
The patent replaces direct mechanical signal comparison with probabilistic modeling using Hidden Markov Models. This substitution transforms the comparison mechanism from deterministic signal matching to statistical distribution comparison, achieving higher detection reliability through robust probabilistic reasoning while managing computational complexity through efficient probability calculations.
3Measurement precision
If time alignment methods are applied to improve acoustic event comparison accuracy, then the measurement precision improves, but the device complexity and computational requirements increase
Solution Approach 1:
The patent introduces probability distributions over subword units as an intermediary between raw acoustic events and final similarity comparison. This intermediary layer enables time alignment and comparison operations to be performed in a standardized probabilistic space, improving measurement precision while simplifying the overall processing complexity through unified mathematical operations.
Data Source
AI summary
An approach to comparing events in word spotting, such as comparing putative and reference instances of a keyword, makes use of a set of models of subword units. For each of two acoustic events and for each of a series of times in each of the events, a probability associated with each of the models of the set of subword units is computed. Then, a quantity characterizing a comparison of the two acoustic events, one occurring in each of the two acoustic signals, is computed using the computed probabilities associated with each of the models.