Voice biometrics is technology that recognises a person by the sound of their voice. A company uses it to confirm who is calling, without asking for a PIN, a date of birth or answers to security questions. It is also called voice authentication or speaker verification. Biometrics means identifying a person from something about their body, such as a fingerprint, the face or the voice.
Every voice is slightly different. The shape of a person's throat, mouth and nose, and the way they speak, change the sound in measurable ways. Voice biometrics software measures these features and stores them as a voiceprint. A voiceprint is a list of numbers that describes one person's voice. It is not a recording, and the voice cannot be played back from it.
It works in two stages. First, enrolment. With the customer's consent, the software records a short sample of the customer's voice, often during a normal call, and creates the voiceprint. Second, verification. On a later call, the software compares the caller's voice with the stored voiceprint. It gives a score for how closely they match. If the score is above a threshold set by the company, the caller is verified.
Consider a retired schoolteacher in Mysuru who calls her bank's helpline. On her first call last year, she agreed to enrol, and the bank created her voiceprint from about 30 seconds of speech. Today she calls to ask about a cheque. In the first 10 seconds of the call, while she explains her question, the software compares her voice with the voiceprint. It gives a match score of 92 out of 100. The bank's threshold is 80. So the agent's screen shows "Verified", and the agent can help her at once. She did not have to answer any security questions.
The score is never certain. A cold, a noisy room or a poor phone line can lower the score of the real customer. A close relative with a similar voice may score higher than expected. So banks combine the voice check with other checks, such as the phone number the call comes from. For risky requests, such as a large transfer, they still ask for an OTP.
A newer risk is cloned voices. Software can now copy a person's voice from a short recording, and fraudsters use such copies. So voice biometrics products add liveness checks, which test whether the voice comes from a live person or from a recording or a machine. This is closely related to deepfake detection.
Voice biometrics is often built into a conversational AI platform, so the same bot that answers the call can also verify the caller. It is one way of checking a person's identity. Checking a new customer's ID documents and face is done by identity verification (IDV). In India, the company needs the customer's consent to create a voiceprint. Today, the Information Technology Act's rules for sensitive personal data require this consent, because they cover biometric data. From 13 May 2027, the main duties of the Digital Personal Data Protection (DPDP) Act apply instead of those rules.