Confusion Network Distributed Representation for Speech Recognition Error Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing class classifiers in spoken dialogue systems face reduced accuracy when handling speech recognition errors, as they are typically trained on error-free data, leading to a mismatch between learning and estimation inputs, and there is a lack of suitable representation methods for confusion networks to be used in machine learning.
Innovation Solution
A confusion network distributed representation sequence is generated by transforming arc word sets and their corresponding weight sets into vector sequences, allowing the confusion network to be used as input for machine learning, which improves the performance of class classifiers by incorporating recognition errors and hypothesis spaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If class classifiers are trained on error-free speech text, then learning accuracy is improved, but estimation accuracy for speech text with recognition errors deteriorates
Solution Approach 1:
The invention applies preliminary action by generating multiple candidate hypotheses (confusion network) before final classification. Instead of directly classifying the single speech recognition result, the system first generates a confusion network containing multiple possible word sequences with different probabilities, then uses this enriched representation as input to the classifier, improving robustness to recognition errors.
Solution Approach 2:
The invention changes the input parameter representation by transforming the single speech recognition result into a confusion network distributed representation. This involves encoding multiple hypotheses with their probabilities into a vector sequence format, fundamentally changing how the classifier receives and processes input information, thereby improving estimation accuracy for texts with recognition errors.
2Reliability
If confusion network is used as input for machine learning, then estimation accuracy is improved, but a suitable representation method was lacking
Solution Approach 1:
The invention substitutes the mechanical/structural confusion network representation with a distributed vector representation. Instead of using the original graph-based confusion network structure, the system transforms it into a sequence of vectors that can be directly processed by standard machine learning models, replacing the complex structural representation with a simpler numerical one.
Solution Approach 2:
The invention changes the representation parameters by converting the confusion network's discrete graph structure into continuous vector sequences. This parameter transformation enables the confusion network to be compatible with standard machine learning input requirements while preserving the essential information about multiple hypotheses and their probabilities.
Data Source
AI summary
There is provided a technique for transforming a confusion network to a representation that can be used as an input for machine learning. A confusion network distributed representation sequence generating part that generates a confusion network distributed representation sequence, which is a vector sequence, from an arc word set sequence and an arc weight set sequence constituting the confusion network is included. The confusion network distributed representation sequence generating part comprises: an arc word distributed representation set sequence transforming part that, by transforming an arc word included in the arc word set to a word distributed representation, obtains an arc word distributed representation set and generates an arc word distributed representation set sequence; and an arc word distributed representation set weighting/integrating part that generates the confusion network distributed representation sequence from the arc word distributed representation set sequence and the arc weight set sequence.