Dynamic Time Warping for Dialog Repeat Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dialog systems struggle to reliably detect repeated user inputs without access to internal speech recognition engine information and fail to recognize repeated phrases with varying word orders, leading to poor user satisfaction and recognition performance.
Innovation Solution
A dynamic-time warping (DTW) based pattern comparison algorithm is used to detect repeated parts between original and correction utterances, employing unsupervised segmentation and recursive DTW computation to identify matching patterns without requiring fixed end-points or exact word order, utilizing spectral characteristics and Mel-frequency cepstral coefficients for feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dynamic time warping (DTW) and N-best hypotheses overlapping measures are used to detect repeated utterances, then the system can identify misrecognized parts of the original utterance, but the order of words or components must be the same as in the original utterance, which limits detection accuracy when users alter word order
Solution Approach 1:
The patent segments the correction utterance into multiple candidate repeated parts using unsupervised segmentation. This allows the system to independently analyze different portions of the correction utterance and identify which segments are repetitions, regardless of their position or order relative to the original utterance. The segmentation enables flexible matching without requiring the entire utterance to follow the same word order.
Solution Approach 2:
The patent transforms the repeat detection problem from a strict sequence-matching approach to a pattern-matching approach by extracting spectral characteristics and Mel-frequency cepstral coefficients. This dimensional transformation allows comparison based on acoustic patterns rather than positional sequence, enabling detection of repetitions even when word order changes.
2Reliability
If present repeat detection systems require access to internal components of the speech recognizer process, then they can perform repeat detection, but such access is often not convenient or even available, further limiting the effectiveness of these systems
Solution Approach 1:
The patent implements a self-service approach where the repeat detection system processes only the audible utterances themselves without requiring access to internal speech recognizer components. The system extracts spectral characteristics directly from the audio signals and performs pattern comparison independently, making the detection process autonomous and accessible without needing internal engine information.
3Device complexity
If the system uses strict sequence matching for repeat detection, then the implementation is simpler, but users can easily alter the order of words while repeating them, which effectively impacts the performance of the speech recognition capability
Solution Approach 1:
The patent changes the parameters used for comparison from positional sequence information to spectral characteristics and Mel-frequency cepstral coefficients. This parameter transformation allows the system to compare utterances based on their acoustic patterns rather than word positions, maintaining detection reliability while handling natural variations in user speech patterns including word order alterations.
Data Source
AI summary
Embodiments of a method and system for detecting repeated patterns in dialog systems are described. The system includes a dynamic time warping (DTW) based pattern comparison algorithm that is used to find the best matching parts between a correction utterance and an original utterance. Reference patterns are generated from the correction utterance by an unsupervised segmentation scheme. No significant information about the position of the repeated parts in the correction utterance is assumed, as each reference pattern is compared with the original utterance from the beginning of the utterance to the end. A pattern comparison process with DTW is executed without knowledge of fixed end-points. A recursive DTW computation is executed to find the best matching parts that are considered as the repeated parts as well as the end-points of the utterance.