Speech Recognition Expected Response Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition systems face challenges in accurately recognizing words with acoustic models that are not well-matched to input audio, leading to incorrect data entry and user frustration, especially when responses contain multiple pieces of information spoken without pausing, as previous solutions discount entire responses due to mismatches in parts.
Innovation Solution
A speech recognition system that compares and adjusts its performance by considering parts of the hypothesis independently, allowing acceptable hypothesis parts to be adapted or accepted without being rejected due to mismatches in other parts, using a flexible expected response/hypothesis comparison and adjusting acceptance thresholds accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire response is discounted due to a mismatch in one part, then the reliability of the speech recognition system is maintained, but the productivity and user efficiency deteriorate due to excessive rejections of partially correct responses
Solution Approach 1:
The patent divides the response hypothesis into multiple parts (e.g., bin location and quantity picked) and evaluates each part independently against the expected response. This segmentation allows the system to accept partially correct responses rather than rejecting the entire response due to a mismatch in one part, thereby improving productivity while maintaining reliability through selective acceptance.
Solution Approach 2:
The patent applies different acceptance criteria to different parts of the response hypothesis based on their individual confidence scores and match status with expected response parts. Each part can be evaluated with its own threshold adjustment, allowing locally optimized decision-making that improves overall system performance without compromising reliability.
2Measurement precision
If the acceptance threshold is adjusted based on the entire response match, then the measurement precision of response accuracy is improved, but the adaptability to partial matches deteriorates
Solution Approach 1:
The patent segments the response evaluation into multiple independent parts, each with its own precision measurement and adaptability assessment. This allows the system to maintain high measurement precision for each part while adapting flexibly to partial matches across different parts of the response.
Solution Approach 2:
The patent dynamically adjusts acceptance thresholds for each part of the response based on individual confidence scores and match status. This dynamic adjustment allows the system to adapt to varying degrees of match quality across different parts, improving both precision and adaptability simultaneously.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A speech recognition system used for hands-free data entry receives and analyzes speech input to recognize and accept a user's response. Under certain conditions, a user's response might be expected. In these situations, the expected response may modify the behavior of the speech recognition system to improve performance. For example, if the hypothesis of a user's response matches the expected response then there is a high probability that the user's response was recognized correctly. This information may be used to make adjustments. An expected response may include expected response parts, each part containing expected words. By considering an expected response as the concatenation of expected response parts, each part may be considered independently for the purposes of adjusting an acceptance algorithm, adjusting a model, or recording an apparent error. In this way, the speech recognition system may make modifications based on a wide range of user responses.