Specification

The BANCA database was designed in order to test multi-modal IV with various acquisition devices (2 cameras and 2 microphones) and under several scenarios (controlled, degraded and adverse) (controlled, degraded and adverse). For 4 different languages (English, French, Italian and Spanish), video and speech data were collected for 52 subjects (26 males and 26 females) on 12 different occasions, i.e. a total of 208 subjects. Each language - and gender - specific population was itself subdivided into 2 groups of 13 subjects, denoted in the following g1 and g2.

Each subject recorded 12 sessions, each of these sessions containing 2 recordings: 1 true client access and 1 informed (the actual subject knew the text that the claimed identity subject was supposed to utter) impostor attack. The 12 sessions were separated into 3 different scenarios:

The web cam was used in the degraded scenario, while the expensive camera was used in the controlled and adverse scenarios. The two microphones were used simultaneously in each of the three scenarios with each output being recorded onto a separate track of the DV tape.

During each recording, the subject was prompted to say a random 12 digit number, his/her name, their address and date of birth. Each recording took an average of twenty seconds. Table 1 gives an example of the speech a subject would be expected to utter at a single session (i.e. two recordings). For each session the true client information remained the same. For different sessions the impostor attack information changed to another person in their group.

True Client

Impostor Attack

        0 3 8 9 2 1 6 7 4 5 0 1         

        8 5 7 9 0 1 3 2 4 6 0 2         

Annie Other

Gertrude Smith

9 St Peters Street

12 Church Road

Guildford

Portsmouth

Surrey

Hampshire

GU2 4TH

PO1 3EF

20.02.1971

12.02.1976

Table 1: Example of the speech uttered by a subject at one of the twelve Banca sessions.

More formally, in a given session, the impostor accesses by subject X were successively made with a claimed identity corresponding to each other subject from the same group (as X). In other words, all the subjects in group g recorded one (and only one) impostor attempt against each other subject in g and each subject in group g was attacked once (and only once) by each other subject in g. Moreover, the sequence of impostor attacks was designed so as to make sure that each identity was attacked exactly 4 times in the 3 different conditions (hence 12 attacks in total).