Use Case Considerations

One of the most critical elements of any voice biometric system is the use case selected. The term "use case" refers to the general prompting technique that the user experiences within an application. The quality of this experience is directly related to user acceptance and long-term viability of the voice biometric system. So, system designers must be careful to weigh all factors before committing to a specific use case.

In general, users can be prompted to speak passphrases, short sentences, numbers, or they can even speak conversationally. VBG's technology allows us to support all common use cases:

Static Text Passphrase (ST).  This use case prompts users to speak a static passphrase, such as "My voice is my password, please let me in." Enrollment typically requires 2-3 repetitions of the same phrase, while verification and identification require the phrase to be spoken once. In theory each user can select his or her own passphrase, but this can be difficult to administer and support in practice. So, companies typically have everyone repeat the same basic passphrase.

Static Numeric Passphrase (SN).  This use case is very similar to text-based passphrase use cases, except that specific numbers are repeated several times to enroll. Also, users typically repeat a number that is unique to them instead of having everyone repeat the same number. As an example, we have several clients that are using cell phone numbers or partial account numbers as numeric phrases for their customers.

Random Number (RN).  With this use case, a series of static number phrases and/or counting phrases are prompted for in order to obtain samples of how users speak each digit (0 through 9). Then, during verification or identification, users are prompted to repeat a random number -- or any other number, typically 4+ digits long. This technique is fairly simple to administer and most users find it easy to work with.

Free Speech (FS).  This use case, sometimes also referred to as "natural speech", requires 1-2+ minutes of speech samples to create a robust voice print. The idea is to capture and model all the phonemes of speech, so that a user can say just about anything later on to verify or be identified (passphrases, numbers, or conversational speech). It is an extremely flexible technique for system designers, but it can require a significant amount of up-front time from users.