Voice Recognition Threshold Tuning for Short Utterance Enrollment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Voice recognition systems struggle to accurately identify users based on shorter utterances, particularly for children and different genders, leading to inaccuracies and discouragement in enrollment for voice identity services.
Innovation Solution
An algorithm processes voice input data to determine characteristics such as gender, age, and accent, adjusting confidence thresholds for improved user identification and command recognition by using implicit and explicit enrollment methods, including unsupervised machine learning for fingerprint generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voice recognition systems use standard algorithms with fixed confidence thresholds, then the system structure remains simple, but accuracy deteriorates for children's voices and different genders
Solution Approach 1:
The system dynamically adjusts confidence thresholds based on detected voice characteristics such as gender and age. Instead of using fixed thresholds, the system adapts thresholds in real-time according to the speaker's profile, improving recognition accuracy for diverse voice types while managing complexity through automated adaptation
Solution Approach 2:
The system changes the confidence threshold parameter based on voice characteristics. By detecting gender and age from the voice input, the system selects appropriate threshold values from a set of predefined thresholds, allowing accurate recognition across different demographic groups without requiring complex manual configuration
2Measurement precision
If voice recognition systems require long and multiple enrollment sentences, then voice identity accuracy improves, but user convenience deteriorates
Solution Approach 1:
The system uses partial enrollment actions by requiring only short utterances rather than long sentences. By combining multiple short enrollments with dynamic threshold adjustment during recognition, the system achieves accurate voice identity verification while significantly reducing the burden on users during enrollment
Solution Approach 2:
The system performs preliminary processing during enrollment to extract voice characteristics and establish baseline profiles from short utterances. This preliminary action enables the system to make accurate recognitions later without requiring users to provide lengthy enrollment samples, improving both accuracy and convenience
3Measurement precision
If voice recognition systems use gender-specific confidence thresholds, then recognition accuracy for different genders improves, but system complexity increases
Solution Approach 1:
The system automatically detects gender and age from voice characteristics and self-adjusts the confidence threshold without manual intervention. This self-service approach improves recognition accuracy for different genders while eliminating the need for complex manual threshold management, as the system adapts thresholds autonomously based on detected voice profiles
Data Source
AI summary
Methods and systems are described for recognizing, based on a voice input, a user and/or a voice command. An algorithm is described herein that processes data associated with a voice input. The data may indicate characteristics of the voice such as a gender, an age, or accent associated with the voice and other metadata. For example, the system may process the data and determine the gender of a voice. The determined characteristics may be used as an input into a voice recognition engine to improve the accuracy of identifying the user who spoke the voice input and identifying a voice command associated with the voice input. For example, the determined gender may be used as a parameter to improve the accuracy of an identified user (e.g., the speaker) or command. The algorithm may adjust, based on gender, parameters such as confidence thresholds used to match voices and voice commands.