Speech Recognition False Alarm Reduction via Multi-Level Context Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic speech recognition systems often experience false alarms due to incorrect detection of spoken words that are acoustically similar to user-defined words, leading to detection errors.
Innovation Solution
The system employs spoken word verification by analyzing global and local contexts on lexical, phonetic, and acoustical levels, generating models from training examples, and comparing test words with these models to determine if they meet a threshold for acceptance or rejection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If speech recognition systems use simple acoustic matching, then detection speed is fast, but false alarm rate increases due to acoustically similar words
Solution Approach 1:
The verification process is segmented into multiple independent context levels: acoustic context, phonetic context, and lexical context. Each level generates separate models that are independently compared against the test word, allowing the system to break down the complex verification task into manageable segments that can be processed sequentially or in parallel.
Solution Approach 2:
The system transitions from single-dimensional acoustic matching to multi-dimensional verification by incorporating phonetic and lexical context dimensions. This dimensional expansion allows the system to distinguish between acoustically similar words by evaluating them across multiple contextual axes simultaneously.
2Measurement precision
If multiple contextual models are generated and compared, then verification accuracy improves, but processing time increases
Solution Approach 1:
Context models are generated in advance during a training phase using stored training examples. This preliminary action allows the system to pre-compute acoustic, phonetic, and lexical context models, so that during actual verification, the system only needs to perform comparison operations rather than generating models in real-time, significantly reducing processing time.
Solution Approach 2:
The system can perform partial verification by evaluating only the most discriminative context levels for a given word pair. If acoustic context alone provides sufficient discrimination, phonetic and lexical contexts may be skipped, reducing processing time while maintaining adequate accuracy for clear cases.
3Reliability
If context models are stored for all possible words, then verification reliability improves, but memory requirements increase
Solution Approach 1:
Instead of storing uniform context models for all possible words, the system stores context models selectively for words that appear in training examples and are relevant to the application domain. Each word's context models are tailored to its specific acoustic, phonetic, and lexical characteristics, optimizing storage efficiency while maintaining verification reliability for recognized words.
Data Source
AI summary
A system and method are presented for using spoken word verification to reduce false alarms by exploiting global and local contexts on a lexical level, a phoneme level, and on an acoustical level. The reduction of false alarms may occur through a process that determines whether a word has been detected or if it is a false alarm. Training examples are used to generate models of internal and external contexts which are compared to test word examples. The word may be accepted or rejected based on comparison results. Comparison may be performed either at the end of the process or at multiple steps of the process to determine whether the word is rejected.