Speech Summary Lattice Optimization via Integer Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition summary technologies often include recognition errors and generate redundant summaries, especially when summarizing long utterances with duplicated content, as they do not effectively utilize the concept of important words and typically rely on 1-best speech recognition results.
Innovation Solution
A summary generating apparatus that converts input utterance sequences into k-best confusion network sequences, generates lattice sequences with candidate words as internal nodes, and uses integer programming to select paths that maximize a coverage score of important words, thereby reducing recognition errors and redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If only the 1-best speech recognition result is used to generate a summary, then the processing complexity is reduced, but speech recognition errors are included in the summary
Solution Approach 1:
The patent segments the speech recognition results into multiple candidate hypotheses (k-best results) rather than treating them as a single unit. By dividing the recognition output into multiple plausible alternatives and evaluating each through lattice structures, the system can select the most accurate path while maintaining manageable processing complexity through structured segmentation of the search space.
Solution Approach 2:
The patent performs preliminary action by pre-processing speech recognition results into lattice structures that organize multiple candidate hypotheses before summary generation. This preliminary organization of k-best results into structured lattices with edges representing word transitions enables more accurate summary generation while controlling complexity through structured preprocessing.
2Device complexity
If traditional sentence extraction or word extraction methods are used for summarization, then the processing is simple, but the summary contains redundancy especially for long utterances with duplicated content
Solution Approach 1:
The patent performs preliminary action by pre-processing speech recognition results into lattice structures that organize multiple candidate hypotheses before summary generation. This preliminary organization enables the system to identify and eliminate redundant information through structured path selection, reducing summary redundancy while controlling processing complexity.
Solution Approach 2:
The patent changes parameters by introducing coverage scores and importance weights for different words and phrases when selecting paths through the lattice structure. By adjusting these parameters during path optimization, the system can prioritize unique and important information while suppressing redundant content, thereby reducing summary redundancy.
3Reliability
If k-best speech recognition results are used instead of 1-best, then speech recognition errors are reduced, but the processing complexity increases
Solution Approach 1:
The patent segments the k-best speech recognition results into structured lattice formats with internal nodes representing candidate words and edges representing transitions. This segmentation organizes the complex k-best data into manageable structures, enabling accurate summary generation while controlling processing complexity through systematic organization of multiple hypotheses.
Solution Approach 2:
The patent substitutes mechanical brute-force processing of k-best results with an optimized lattice-based path selection system. By replacing direct examination of all k-best combinations with structured lattice traversal and integer programming optimization, the system achieves high reliability in reducing speech recognition errors while managing processing complexity through mathematical optimization.
4Device complexity
If traditional summarization methods are used without important word coverage concept, then the implementation is straightforward, but the summary has high redundancy
Solution Approach 1:
The patent changes parameters by introducing coverage scores and importance weights for different words when selecting paths through the lattice structure. By adjusting these parameters during optimization, the system prioritizes unique and important information while suppressing redundant content, thereby reducing summary redundancy with moderate implementation complexity.
Data Source
AI summary
A speech recognition unit converts an input utterance sequence into a confusion network sequence constituted by a k-best of candidate words of speech recognition results; a lattice generating unit generates a lattice sequence having the candidate words as internal nodes and a combination of k words among the candidate words for an identical speech as an external node, in which edges are extended between internal nodes other than internal nodes included in an identical external node, from the confusion network sequence; an integer programming problem generating unit generates an integer programming problem for selecting a path that maximizes an objective function including at least a coverage score of an important word, of paths following the internal nodes with the edges extended, in the lattice sequence; and the summary generating unit generates a high-quality summary having less speech recognition errors and low redundancy using candidate words indicated by the internal nodes included in the path selected by solving the integer programming problem, under a constraint on the length of a summary to be generated.