Voice User Identification via Metadata Subset Segmentation

The patent segments the large set of saved audio files into multiple subsets based on metadata characteristics (such as time of recording, device type, or other organizing criteria). The backend then compares the transmitted audio file against multiple smaller subsets in parallel or sequential stages, rather than comparing against all saved audio files at once. This segmentation maintains identification accuracy by ensuring the correct user's audio files are included in the search subsets while dramatically reducing the computational burden and response time.

The patent performs preliminary actions during the user registration phase by collecting and analyzing multiple audio files from each user, establishing baseline voice characteristics and creating comprehensive user profiles. This preliminary action builds a robust reference database that enables the system to reliably authenticate users based solely on voice input during subsequent login attempts, eliminating the need for additional verification inputs while maintaining high security standards.

The patent implements a dynamic subset determination process where the backend initially creates subsets based on metadata analysis to improve processing efficiency. However, the system dynamically adjusts the search process by incorporating flexibility in metadata matching criteria and allowing for partial matches. If initial metadata-based subset filtering misses potential matches, the system can expand the search scope or adjust metadata thresholds, thereby maintaining search accuracy while preserving the productivity benefits of subset-based processing.