Voice biometrics is often synonymous with speaker verification; a one-to-one comparison designed to prove someone says they are who they say they are. However, it plays an equally important role in speaker identification.
At first glance these may seem like different names for the same thing, but there are some significant differences. From a purely technical perspective, verification compares a speaker’s voice to a single reference file (voiceprint) and generates a simple yes/no result.
Identification compares a speaker’s voice to a database of multiple voiceprints and returns a unique value, with a degree of confidence. Of course, if the speaker is not already enrolled in the system, this can be flagged.
The applications for speaker identification are many and varied and can be used to identify a caller in either a self-service or agent-assisted interaction. Perhaps the most obvious application is to identify a caller in an inbound contact centre engagement. If the caller has enrolled in the system, identification can be confirmed rapidly.
The time taken to identify a caller is dependent upon the size of the reference database, so a simple qualifying question can be used to narrow the search parameters. For example, asking the caller for their postcode significantly reduces the overall population. From a practical perspective, voice biometrics is used to identify a caller if they do not have their unique reference to hand.
Outside the traditional contact centre application, speaker identification has a role to play in the emergency services arena. There are a number of known nuisance callers who, for reasons only apparent to themselves, feel compelled to call the emergency services with false alarms. Serial offenders, these nuisance callers are profiled using previous recordings and can be identified in real-time.
Voice biometrics has also been adopted within the prison system to ensure compliant use of prison phone systems. Speaker ID can be used at the outset to ensure the caller has permission to use the phone. It can also be used to ensure that the person who instigated the call is the person who remains on the call, or if the call is passed to a third party.
Not all speaker ID applications take place in real-time. The technology can also be used during forensic audio analysis to identify individuals in recordings. For example, if an existential threat is issued by an unknown organisation, speaker ID can be used to compare the voiceprint against known “persons of interest” to identify the caller.
Speaker identification is a powerful tool for fraud detection and prevention. When dealing with account access or financial transactions, speaker ID provides technical support to agents who may suspect a caller is attempting to pass themselves off as someone they are not.
Speed of identification is a key benefit. In the midst of a conversation, a positive identification can be made in a matter of seconds (for smaller databases). If the reference list is significant (millions of potential results), speech recognition can be combined with speaker ID to narrow the search criteria. This in-call identification can have a significant impact on average call duration and provides a frictionless customer experience.
If you are thinking of introducing voice biometrics for identification in your business, and would like to discuss your requirements, contact one of our consultants today.