Menu

What the stus of ALIZE ?

2015-03-24
2015-03-28
  • Travis Banger

    Travis Banger - 2015-03-24

    Note: I do not intend to use ALIZE like this:

    "Abracadabra! Hocus Pocus"

    [and the door, trained to my voice, opens]

    What I need is to submit a set of voice clips known to be spoken by a subject (Mr. Bill O'Reilly) and determine whether the voice in the target voice clip is indeed, his.

    Those two tasks described above seem equivalent, right?

    We are dealing with Speaker Recognition, correct?

     
    • Travis Banger

      Travis Banger - 2015-03-25

      We are dealing with Speaker Recognition, correct?

      Indeed, my esteemed alter ego! However, Speaker Recognition is subdivided in two distinct categories (actually, more):

      • Speaker Identification ("Abracadabra, I command you to open the door!")

      • Speaker Verification (is the speaker who he claims -or not!- to be?)

      Here there is a highly readable paper about this:

      "Speaker Recognition - Identifying People by their Voices"
      1.2 Areas of Automatic Speaker Recognition
      http://goo.gl/wnn8R4

       
  • Travis Banger

    Travis Banger - 2015-03-24

    I found the ALIZE mailing list and subscribed.

    My question remains open. I'd like to make sure that I am going in the right direction.

    TIA

     
    • Nickolay V. Shmyrev

      My question remains open. I'd like to make sure that I am going in the right direction.

      Alize is one of the popular speaker identification toolkits

       
      • Travis Banger

        Travis Banger - 2015-03-28

        Alize is one of the popular speaker identification toolkits

        I'd give anything to see a formal comparison between Alize and this state-of-the-art, commercial product:

        https://www.youtube.com/watch?v=5EHnNYwgo-U

         
        • Nickolay V. Shmyrev

          Alize does the job automatically, this video demonstrates human investigation.

          NIST holds a number of challenges to evaluate speaker recognition performance.

          SpeechPRO participated in a latest speech identification competition, you can find results here:

          http://cs.uef.fi/odyssey2014/program/pdfs/25.pdf

          Alize results are covered here:

          http://www1.i2r.a-star.edu.sg/~alarcher/Publications_files/ALIZE_3.pdf

          The algorithms implemented in both systems are similar, the difference is within 50%, it is not by order of magnitude.

           
          • Travis Banger

            Travis Banger - 2015-03-28

            [Nick:]

            Alize does the job automatically, this video demonstrates human investigation.

            I was so naive, that I thought all I had to do was manually (*) normalize the same utterance from the reference/known person and the unknown person. Let's say, make sure that the range of amplitude and frequencies were the same, would superimpose the 2 images and it would be immediately obvious whether the formats were equivalent.

            For starters, I have learned that instead of using a sound-generic application, I should use one specialized in voice, like Praat. Interestingly, there is an Alize plugin for Praat.

            (*) I use:

            • iZotope RX4
            • Adobe Audition
            • Audacity
             

            Last edit: Travis Banger 2015-03-28
  • Travis Banger

    Travis Banger - 2015-03-24

    To those folks out there interested in ALIZE, the best resource (with questions and answers) is LinkedIn:

    "Alize Biometry platform"

    https://www.linkedin.com/groups?home=&gid=2323703&trk=anet_ug_hm

     

Log in to post a comment.