Neural Network Ranker for Hybrid Speech Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems that utilize multiple engines face challenges in accurately selecting the best speech recognition result from a set of candidate outputs, as low-level combinations of outputs fail to preserve high-level linguistic information and determining the optimal result is complex.
Innovation Solution
A method involving a controller that generates feature vectors for candidate speech recognition results, identifies trigger pairs, and uses a neural network to produce ranking scores, selecting the result with the highest score for operation, and a training process for the neural network ranker using feature vectors and target results based on edit distances to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple speech recognition engines are used to generate candidate results, then speech recognition accuracy is improved, but the complexity of selecting the optimal result increases
Solution Approach 1:
The patent introduces a neural network ranker as an intermediary component that mediates between multiple speech recognition engines and the final selection process. The ranker takes candidate results from multiple engines, processes them through learned features, and outputs ranked scores that guide the selection of the optimal result, thereby simplifying the overall decision-making process while maintaining high accuracy
Solution Approach 2:
The system transforms speech recognition results into numerical ranking scores through the neural network, changing the parameter representation from unstructured text to structured numerical values. This parameter transformation enables efficient comparison and ranking of multiple candidate results, reducing the complexity of selecting the optimal outcome
2Device complexity
If low-level combinations of outputs from multiple speech recognition systems are used, then processing simplicity is maintained, but high-level linguistic information is lost
Solution Approach 1:
The system performs preliminary processing by generating complete speech recognition results from multiple engines before combining them, rather than combining low-level acoustic features first. This preliminary generation of full results preserves high-level linguistic information while allowing subsequent neural network processing to efficiently rank and select the best outcome
Data Source
AI summary
A method for ranking candidate speech recognition results includes generating, with a controller, a plurality of feature vectors for the candidate speech recognition results, each feature vector including one or more of trigger pair features, a confidence score feature, and word-level features. The method further includes providing the plurality of feature vectors as inputs to a neural network, generating a plurality of ranking scores corresponding to the plurality of feature vectors for the plurality of candidate speech recognition results based on an output layer of the neural network, and operating the automated system using the candidate speech recognition result in the plurality of candidate speech recognition results corresponding to a highest ranking score in the plurality of ranking scores as input.