Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

Dynamic Test Durations for Text-Independent Speaker Verification Systems

Axel Glaeser

Ascom AG, Applicable Research & Technology Unit, Mägenwil, Switzerland

In the past few years, phone banking and phone shopping have become more and more popular. These new needs have mainly been initiated by the service providers (banks, etc…) which are looking for cost-efficient and secure solutions to enable their clients a flexible and comfortable service access. In this context, the assessment of a speaker recognition system must take into account subjective factors that are directly linked to user acceptance criteria. During field tests we found out that the main point of criticism was the test duration. While the enrollment duration was mainly accepted as a single, initial effort, the test duration of 10 seconds was not accepted as feature of a user-friendly system. Therefore, we developed a new approach based on Dynamic Test Durations (DTD) by taking the quality of speech and special characteristics of the classification process into account. This approach results in a decrease of the test down to 2,4 seconds (average).


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Glaeser, Axel (1999): "Dynamic test durations for text-independent speaker verification systems", In EUROSPEECH'99, 1199-1202.