Voice Biometric Accuracy via Audio Segmentation and Noise Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Voice biometrics face challenges in obtaining a clear user voice sample due to background noise and inconvenient quiet environment requirements, making it difficult to generate accurate biometrics from call audio.
Innovation Solution
A biometric system uses machine learning to differentiate between user and agent voices in call audio recordings by generating signature vectors and comparing them using a distance metric, removing non-matching segments to create a more accurate voice biometric.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is used to process call audio to generate voice biometrics, then the accuracy of voice biometric generation is improved, but the complexity of the system increases
Solution Approach 1:
The call audio is divided into multiple segments, with the first segment used to generate a signature vector and subsequent segments compared against it. This segmentation allows the system to process audio in manageable parts, improving accuracy through systematic comparison while managing computational complexity through modular processing.
Solution Approach 2:
The system performs preliminary processing by generating a signature vector from the first audio segment before comparing it with subsequent segments. This preliminary action establishes a reference point that simplifies subsequent processing, as the system only needs to compare rather than fully analyze each segment, balancing accuracy with computational efficiency.
2Quantity of substance
If the system processes the entire user audio channel including transferred agents, then more voice data is available for biometric generation, but the reliability of the biometric decreases due to mixed voices
Solution Approach 1:
The system extracts and isolates the user's voice from the mixed audio channel by comparing segments against a signature vector generated from the user's initial speech. Segments that match the signature are extracted for biometric generation, while non-matching segments (containing agent voices or noise) are excluded, ensuring high reliability while utilizing available voice data.
Solution Approach 2:
Different segments of the audio channel are treated differently based on their quality and source. The system identifies segments containing the user's voice through signature matching and processes them with high priority, while segments containing agent voices or background noise are handled separately or excluded, ensuring that the biometric is generated from high-quality, reliable data.
3Measurement precision
If the system requires users to go to quiet environments for voice sampling, then the quality of voice biometric is improved, but the ease of operation decreases
Solution Approach 1:
The system converts the harmful effect of background noise and mixed voices in call audio into a beneficial processing opportunity. By using machine learning to analyze and separate voices in the audio channel, the system can extract clean user voice samples even from noisy environments, eliminating the need for users to seek quiet locations while maintaining high biometric quality.
Solution Approach 2:
The system replaces the mechanical requirement of physical quiet environments with an intelligent audio processing system. Instead of requiring users to physically relocate to quiet spaces, the machine learning system processes the audio digitally, using signature vector comparison to isolate the user's voice from background noise and other speakers, achieving the same quality goal through computational rather than physical means.
Data Source
AI summary
Methods and systems are disclosed herein for improving the quality of audio for use in a biometric. A biometric system may use machine learning to determine whether audio or a portion of the audio should be used as a biometric for a user. A sample of the user's voice may be used to generate a voice signature of the user. Portions of the audio that do not meet a similarity threshold when compared with the voice signature may be removed from the audio. Additionally or alternatively, interfering noises may be detected and removed from the audio to improve the quality of a voice biometric generated from the audio.