Speech Verification

This system is some work done out of Rice University and implements a Simple Verification Process after speaker has trained system with password.

Data collection

The person speaks a password into the microphone.

Endpoint Detection

The short-time magnitude of the signal is found and recorded, as is the pitch track.

Rate Adjustment to Speaker's Stored Password Signal

Each is cropped and dynamically time warped so that (possible) corresponding points on the signals are aligned.

Filtering to Compare Input Signal and Speaker's Stored Password Signal

Now that the signals rest on top of each other, a matched filter (for both magnitude and pitch) to determine a numerical value for their correlation. These numbers are compared to the thresholds set when the key was first created. If both the magnitude and pitch correlations are above this threshold, the speaker has been verified. Allow user to enter top-secret hangout.

 

Results

on a very SMALL sample (4 people):

  • positive recognition rate: 60%-80%
  • false recognition rate: 8-24% (with knowing password)
  • false recognition rate: 0% (without knowledge of password)

 

 
© Lynne Grewe