Voice Biometrics
Overview
Voice biometrics (speaker recognition) identifies a person by the unique characteristics of their voice — shaped by the size and shape of their vocal tract, learned speech patterns, and accent. This module demonstrates the standard MFCC-based approach used in telephone banking and smart speaker authentication.
What is measured
The browser records a short audio clip and performs short-time Fourier analysis to produce a spectrogram. From the spectrogram, Mel-Frequency Cepstral Coefficients (MFCCs) are extracted per frame:
| Feature | Description |
|---|---|
| MFCC vector | 13 coefficients per frame capturing the spectral envelope of the voice |
| Mean MFCC | The mean across all frames — forms the enrolment profile |
| Pitch | Fundamental frequency estimate, displayed for visualisation |
MFCCs approximate how the human auditory system perceives sound, making them effective for speaker characterisation.
Enrolment
The student makes the configured number of recordings (default: 3) each lasting the configured duration (default: 10 seconds). After each recording the page shows progress dots and prompts for the next attempt. Once all recordings are collected, the mean MFCC vector from each is averaged into a single representative profile.
Identification
A second recording of the same duration is taken. Its mean MFCC vector is compared against all enrolled profiles using cosine similarity. Similarities are normalised to confidence percentages. See How It Works for the formula.
Configuration
All settings are managed via the Admin Panel — no code changes required.
| Setting | Default | Description |
|---|---|---|
| Recording duration | 10 s |
How many seconds of audio are captured per recording. Range: 3–60 seconds. Longer recordings produce a more stable MFCC profile but take more time per student. |
| Enrolment attempts required | 3 |
How many voice recordings to collect during enrolment. Each recording's mean MFCC is averaged with the others to form the final profile. Range: 1–10. |
Requirements
- A microphone and browser permission to access it
- A reasonably quiet environment — background noise degrades accuracy significantly
Discussion points
- Ask students to try different voices (whisper, high pitch) and observe the match score.
- What environmental factors degrade voice recognition?
- Where is voice biometrics used commercially? (Phone banking, Alexa voice profiles, etc.)
- What is a "replay attack" and how would you defend against it?