Speech Recognition Candidate Selection via Dynamic Rule Execution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition technologies face challenges in real-time selection of target results due to the time-consuming calculation of perplexity using RNN language models, leading to poor real-time performance.
Innovation Solution
The method employs a set of selection rules with varying execution sequences, prioritizing those with lower algorithm complexity to quickly identify the target result from multiple candidate recognition results, rather than relying on perplexity calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If perplexity calculation using RNN language model is used to select target result, then recognition accuracy is improved, but real-time performance deteriorates due to long calculation time
Solution Approach 1:
The patent segments the candidate recognition results into multiple groups based on initial scoring, and applies different selection rules to different groups. This segmentation allows the system to avoid exhaustive perplexity calculation on all candidates while maintaining accuracy for the most promising results.
Solution Approach 2:
The patent changes the selection parameters by introducing multiple selection rules with different complexity levels (e.g., rule-based filtering, probabilistic selection, and perplexity calculation) and dynamically choosing which parameter to optimize based on the candidate's characteristics and system state.
2Productivity
If multiple selection rules with different complexities are used, then real-time selection efficiency is improved, but device complexity increases
Solution Approach 1:
The patent implements dynamic selection of selection rules based on runtime conditions such as candidate quality, time constraints, and confidence levels. The system adapts its complexity dynamically, using simpler rules when sufficient and reserving complex perplexity calculation for cases where it provides marginal gains.
Solution Approach 2:
The patent applies partial action by using simplified selection rules for most candidate results and reserving the computationally expensive perplexity calculation only for a subset of high-priority candidates, thereby achieving sufficient accuracy without excessive computational effort.
Data Source
AI summary
A speech recognition method is provided. The method includes: obtaining a voice signal; processing the voice signal according to a speech recognition algorithm to obtain n candidate recognition results, the candidate recognition results including text information corresponding to the voice signal; identifying a target result from among the n candidate recognition results according to a selection rule selected from among m selection rules, the selection rule having an execution sequence of j, the target result being a candidate recognition result that has a highest matching degree with the voice signal in the n candidate recognition results, an initial value of j being 1; and identifying the target result from among the n candidate recognition results according to a selection rule having an execution sequence of j+1 based on the target result not being identified according to the selection rule having the execution sequence of j.