Attribute Identification Device Rejecting Low-Reliability Speech Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing attribute identification technologies face challenges in differentiating uttered speech from television noise, leading to erroneous identification due to varying noise conditions, which affects usability by providing unreliable results.
Innovation Solution
An attribute identification device that calculates posteriori probabilities and reliability of speech frames, rejecting attribute identification results if the reliability falls below a predetermined threshold to prevent impairment of usability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If learning is performed by using learned data limited to some noise conditions, then the attribute identification can be performed efficiently, but speech included in noise is learned as a feature for attribute identification which causes error in identification under low-noise conditions
Solution Approach 1:
The speech signal is segmented into multiple frames, and posteriori probabilities are calculated for each frame independently. This allows the system to evaluate reliability at the frame level and reject only low-reliability frames rather than discarding all results, thus maintaining efficiency while improving accuracy.
Solution Approach 2:
The system changes the parameter of reliability evaluation by introducing posteriori probability calculation and reliability thresholds. By dynamically adjusting which frames are accepted based on their reliability scores, the system can adapt to different noise conditions without retraining, resolving the contradiction between efficiency and accuracy.
2Reliability
If the variety of noise conditions is considered exhaustively, then robust operation under any noise conditions can be achieved, but exhaustive learning becomes impossible due to the wide variety of noise conditions
Solution Approach 1:
The system performs self-evaluation of its own confidence through posteriori probability calculation. By automatically assessing the reliability of each identification result and rejecting low-confidence results, the system achieves robustness without requiring exhaustive training data for all possible noise conditions.
Solution Approach 2:
Instead of changing the training data to cover all noise conditions, the system changes the evaluation parameter by introducing reliability scores. This allows the same trained model to adapt to various noise conditions dynamically, avoiding the complexity of exhaustive learning while maintaining robustness.
3Ease of operation
If attribute identification results are given in a uniform manner, then the system operates simply, but usability is impaired due to provision of erroneous identification results
Solution Approach 1:
The system dynamically adjusts its output behavior based on reliability assessment. Instead of uniformly accepting all results, it adaptively rejects low-reliability frames while accepting high-reliability ones. This dynamic approach maintains simplicity by using a single reliability threshold mechanism while significantly improving usability by filtering out erroneous results.
Data Source
AI summary
An attribute identification technology that can reject an attribute identification result if the reliability thereof is low is provided. An attribute identification device includes: a posteriori probability calculation unit 110 that calculates, from input speech, a posteriori probability sequence {q(c, i)} which is a sequence of the posteriori probabilities q(c, i) that a frame i of the input speech is a class c; a reliability calculation unit 120 that calculates, from the posteriori probability sequence {q(c, i)}, reliability r(c) indicating the extent to which the class c is a correct attribute identification result; and an attribute identification result generating unit 130 that generates an attribute identification result L of the input speech from the posteriori probability sequence {q(c, i)} and the reliability r(c). The attribute identification result generating unit 130 obtains a most probable estimated class c{circumflex over ( )}, which is a class that is estimated to be the most probable attribute, from the posteriori probability sequence {q(c, i)} and sets ϕ indicating rejection as the attribute identification result L if the reliability r(c{circumflex over ( )}) of the most probable estimated class c{circumflex over ( )} falls within a predetermined range indicating that the reliability r(c{circumflex over ( )}) is low and sets the most probable estimated class c{circumflex over ( )} as the attribute identification result L otherwise.