Speaker Authentication Operational Test and Evaluation.
Abstract
The long term goal of this program is to implement a realtime speaker recognition system that can be operated by untrained personnel. The objective of the present contract was to demonstrate the feasibility of using speaker recognition techniques to aid in the identification of unknown speakers. The contract was conducted in two phases. The first was the algorithm selection phase, in which various speaker recognition techniques were implemented and tested to determine the optimal technique to satisfy the specific program requirements. The second phase of the program was to implement the selected technique in a laboratory demonstration system, and to develop a convenient, easy to use operator interface to the system. The most difficult task was to develop a text independent speaker recognition technique that would achieve a high level of recognition accuracy with only 10 seconds of training data for each talker. Previous text independent speaker recognition techniques had been reported that had achieved the required accuracies, but all had used several minutes of training data to generate the speaker reference models. In addition, recognition was accomplished in most of these studies by using at least one minute of unknown speech.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1980
- Accession Number
- ADA085845
Entities
People
- Edwin H. Wrench Jr