Chapter 7. Voice Biometric Technologies

The use of voice for biometric authentication seems natural and appropriate. From an early age, we learn to recognize the sound of our parents' voices. Voice provides a very important source of sound for the sense of hearing. The interpretation of a voice we perceive can tell us a lot about someone. It can tell us their relative distance to us, their emotion, and, most importantly, we can associate the voice with someone we know. Like recognizing a face, we have all made the mistake of thinking that we recognized a voice when we did not. This could have happened due to the acoustic conditions under which we heard the voice. In addition, our sense of what direction and how far someone is away from us can also be fooled. To try this, have someone stand behind you and speak to you from a given side. Then take a paper tube and hold it up to the ear that is closest to the person. Have that person speak again. Surprisingly, the person now seems to be closer on the other side. That is because the sound waves need to travel farther to reach the ear. Thus, in measuring the time it takes the sound to be heard in both ears, the brain is tricked into thinking the person is actually closer to the other side.

This demonstrates that even the most complex machine, the human brain, can be fooled by sound and, in turn , by the human voice. If it is possible for the brain to be fooled, it seems only logical that a voice biometric can be fooled as well.

Not surprisingly, the use of voice biometrics for identification is often questioned. Can it deliver the same level of accuracy as other biometrics, and can it be deployed for use in a network security environment?

In our investigation of voice biometrics, we will try to answer these questions and provide a suitable background so that you can make the final decision.

To learn the most about voice biometrics, the discussion will be broken into the following areas:

• General description of voice biometrics

• How is the voice captured?

• Types of algorithms used for voice interpretation

• How can this biometric be spoofed?

General Description of Voice Biometrics

When we speak, words can be broken down into individual components known as phonemes. Each phoneme is delivered with pitch, cadence, and inflection . These three aspects of voice give each of us our unique voice sound. Even though we are all unique individuals and speak for ourselves , we can all sound very similar. This similarity can come from cultural and regional influences in the form of an accent ; also, we learn certain inflections and pitches from our family. That is, if you spend enough time around someone, you can take on some of his/her voice characteristics. Not surprisingly, we have all at one time or another thought we heard one person speak when it was another.

The voice is a physiological and behavioral biometric. That is, it is influenced by our body and the environment we are in. For example, as children grow and pass through puberty, the voice changes. As we grow older, the voice takes on other nuances . Additionally, the sound of our voice can be different if we are talking in a large opera hall or a phone booth , as both have a very unique set of acoustics that will affect the sound of our voice. Therefore, it is possible that our voice does not always sound exactly the same. This again raises the question of the use of voice as an acceptable biometric for network security.

As discussed, the voice can change based on physiological factors and also environmental factors. Thus, both need to be taken into account when evaluating voice-based biometrics.