1

I am working on a project in which the system has to allow only the right person to access the vault. How can I make the system to understand and differentiate a recorded voice and a real human's voice?

7
  • can you specify what is your equipment please?
    – JSmith
    Nov 11, 2015 at 20:12
  • I have a signal generator, an op-amplifier, and a hogh frequency speaker. Now, I have to use a board intel galelio to process the signal and understand it.I also have a power amplifier and a microphone.
    – Code64
    Nov 12, 2015 at 17:24
  • This is a difficult task. I believe the template voice, and the specimen would have to be acquired and compared at a high resolution if the recorded voice was good quality and played back through a full frequency accurate speaker, or faked through voice conversion. The University of Alabama is working on this problem.
    – n00dles
    Nov 12, 2015 at 23:33
  • 1
    @jonhatan -I think Code64 is proposing a security vault with voice authentication (instead of a password), that denies access for a recorded voice (fraudulent access). Banks around the world currently have this problem(well, very similar). It's an interesting question.
    – n00dles
    Nov 13, 2015 at 13:26
  • Playing back a recorded voice requires some kind of speaker, most likely part of a portable system. Most portable speakers have a fairly limited frequency response. That means one place to look for a way to tell live versus recorded would be high frequencies. If live human speech produces much information above say 18 kHz, it's unlikely that a recording would have that. If you can design an ultrasonic microphone into the security system and there is ultrasonic information in live speech, that might work. Nov 13, 2015 at 19:18

0