Keyword Detection Accuracy via Dual-Threshold Noise Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech keyword detection systems face challenges in accurately distinguishing keywords from background noise, leading to false detections and reduced accuracy.
Innovation Solution
An information processing apparatus and method that calculates both keyword and background noise scores using a trained model, determining the presence of a keyword based on the scores and thresholds, thereby reducing false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech keyword detection is performed using conventional methods, then the system can identify keywords from speech, but false detections occur due to inability to distinguish keywords from background noise
Solution Approach 1:
The detection process is segmented into two independent scoring operations: one for keyword detection and another for background noise detection. The keyword score calculation unit computes a score indicating likelihood of keyword presence, while the background noise score calculation unit computes a separate score indicating likelihood of background noise. This segmentation allows independent optimization of each detection pathway and enables comparative analysis to resolve false detections.
Solution Approach 2:
A determination unit acts as an intermediary that receives both the keyword score and background noise score, along with their respective thresholds, and integrates this information to make the final detection decision. This intermediary component compares the keyword score against its threshold and the background noise score against its threshold, then combines these results to determine whether a keyword is truly present, thereby mediating between the two detection signals to eliminate false positives.
2Device complexity
If a single score threshold is used for keyword detection, then the detection process is simple, but accuracy decreases due to inability to account for background noise variations
Solution Approach 1:
The single threshold approach is segmented into two separate threshold mechanisms: a keyword score threshold and a background noise score threshold. Each threshold is optimized independently for its specific detection purpose. The keyword score threshold determines when keyword-like patterns are detected, while the background noise score threshold determines when background noise patterns are detected. This segmentation enables more precise control over detection accuracy without excessive complexity.
Solution Approach 2:
The system changes the detection parameter from a single score threshold to a dual-threshold system with distinct threshold values for keyword and background noise scores. This parameter change allows the system to account for the different statistical characteristics of keyword speech and background noise, improving detection accuracy by using appropriately tuned thresholds for each type of signal rather than applying a uniform threshold that cannot adapt to varying conditions.
Data Source
AI summary
According to one embodiment, an information processing apparatus include following units. The first acquisition unit acquires speech data including frames. The second acquisition unit acquires a model trained to, upon input of a feature amount extracted from the speech data, output information indicative of likelihood of each of a plurality of classes including a component of a keyword and a component of background noise. The first calculation unit calculates a keyword score indicative of occurrence probability of the component of the keyword. The second calculation unit calculates a background noise score indicative of occurrence probability of the component of the background noise. The determination unit determines whether or not the speech data includes the keyword.