Voice print recognition for security
Introduction
Voice recognition is the term used to describe recognising a user by the characteristics of their voice and also comes under the umbrella term of 'biometrics', or using something unique about an individual's physical make-up to identify them. Using voice has become an increasingly popular data input device. There are a number of applications, one of which is in security.
Speaker authentication (verification)
If a user says that they are Mr Smith over the phone and their data of birth is 01/01/1980, than they are claiming to be that person. Authentication, also known as verification, is the process of confirming someone is who they say they are. This can be achieved by using voice technology. A recording of a user's voice is kept in a database and was set up when a user first created an account. When someone claims to be a certain person over the phone, their voice 'print' is retrieved. A sample is taken while they are on the phone and compared to the voice print. This will confirm that they are who they say they are, or will flag up a potential problem that requires further security checks.
Speaker identification
This is similar but different to authentication. Speaker identification is where a voice print is taken over the phone, for example, and a database is searched to try and find a match to the voice print. If a match can be found then the user can be identified. Often, a perfect match cannot be found, although near-matches can. The list of near-matches can be used to quickly identify the user by running further checks on this much smaller number of matches.
Technology
Voice prints can be processed in a number of ways to try and find a match. The simplest method involves a straight forward pattern matching approach, where a person's voice is mapped across to a stored pattern. If enough points coincide, then a match can be confirmed. Sohpisticated maths techniques are also used to identify matches. These include looking at the statistics of the variations in the pitch of the voice and using neural networks, which build up a map of points in the voice print and uses that in a comparison algorithm.
Problems
Problems can occur if there is too much background noise (also called 'ambient noise') when the original voice print was taken, or when a sample is taken for comparison to the original voice print. The effect of this can be reduced using noise reduction algorithms but are not always successful. People change and so do their voices. They may be ill, have a cold or have gained or lost weight. They may be speaking from abroad and the quality of the phone line may not be so good or there may be time delays in the speech. Users may be using different phones at different times, for example, land lines and mobile phones, and they may be using different handsets that have different quality microphones. Any of these factors may have an impact on both the original voice print taken for the database and any subsequent samples taken for verification and identification.
Applications
Many companies use voice recognition to verify users over the phone, including banks and building societies. This reduces the cost of employing people over time and speeds up the processing of enquiries. The claim is that an individual can have their voice sampled and can then be picked out from a database of millions of voice prints. Voice recognition has also been used in the fight against crime and terrorism. Voice prints are taken of terrorists posting speeches on YouTube and to TV stations after carrying out horrible crimes, for example, so that they can be positively linked to them if caught in the future.