One would expect a simple system with no modifications to correctly
identify the speaker 50% of the time. In other words, the system would
either answer yes or no and be correct in half of all occurrences.
With the addition of speech characteristic recognition like pitch and magnitude, one would hope:
To test the system, the four members of the group (JP, Sara, Nipul, Aamir) were set as keys during runs for the system. The basic results are as follows:
Clearly, the use of pitch and magnitude has increased accuracy of the system. However, the system is not perfect. In fact, one must wonder what results in false matches for intruders with the password. The following data sheds some light:
As expected, females were more likely to match Sara and males were more likely to match JP, Nipul, and Aamir.
Also noteworthy is the identity of the intruder in each person's case. A person with multiple intrusions may have characteristics that correlate well with a key.
This correspondence does indicate that similarities in projection of the word could create verification problems.
On a side note, a sociological theory from Nipul gains credence with this data. The multiple matches mentioned above happen to be the roommates of the members of the group. Amazingly, Nipul's theory of vocal absorption states:
"Persons living with other persons gain the latter's vocal characteristics in an exponential relationship that asymptotically approaches k, the annoyance factor, as t approaches infinity."
FUTILITY - "You'll always miss 100% of the shots you don't take, and statistically speaking, 99% of the shots you do."