Voice biometrics

Recognizing someone by how their voice sounds.

For example, a bank's phone system might confirm a caller's identity by comparing their voice with a recording made when they enrolled, without asking them to recite a password.

Speaker recognition from vocal-tract and behavioral features in an audio sample. Vocal-tract features come from the physical shape of a person's throat, mouth, and nasal passages, and behavioral features come from speaking habits such as rhythm and pitch.

A voice biometric system typically works like this:

1. The person enrolls by speaking, and the system converts the audio into a voiceprint, a mathematical summary of their voice.
2. Later, the person speaks again.
3. The system compares the new sample with the stored voiceprint and produces a score.
4. The score is compared with a threshold to accept or reject the match.

Systems use two main approaches:

• Text-dependent systems ask the person to say a fixed phrase.
• Text-independent systems recognize the speaker from any natural speech, which suits phone calls.

Speaker recognition is different from speech recognition. Speaker recognition identifies who is talking, and speech recognition works out what words were said.

Voice biometrics sit next to terms that are easy to mix up:

• Biometrics is the broader field that includes face, fingerprint, and voice.
• Voice cloning is the attack that recreates a person's voice with AI.
• Deepfake is the wider category of AI-generated media that includes synthetic voices.

For more, see Top 5 Cases of AI Deepfake Fraud From 2024 Exposed.