Skip to main content

View Diary: Zimmerman Jury Selection Begins (118 comments)

Comment Preferences

  •  Certain methodologies for voice recognition (2+ / 0-)
    Recommended by:
    Tonedevil, RobertSF

    and speech recognition are accepted by the scientific community and are both widely used, not only in forensics but in commercial applications. Siri is using speech recognition. Voice recognition is used as a security system for securing access to places or information, where a 5 % false positive or false negative is considered acceptable.

    The frye hearings were a masturbatory ego festival for scientists, where they got to insult their collegues and make statements about how the methodologies "ought" to be applied, not about the validity of the methodologies.

    If your standard is 100% positive recognition of every word spoken, you can accuse your collegues for using a substandard sample. But for forensic purposes, that is not necessary. If the sample allows recognition of 50% of the words spoken in a forensic setting, that is valuable information.

    Same for voice recognition. A short sample or one of less than perfect quality, may be not good enough for a security system to establish if the person speaking should be given access to a place or data. But in a forensic setting where the voice heard on the sample, only needs to be matched to 2 known voices, a 90 % match with one of the known persons and 70 % non match for the other, is significant.
    I made those numbers up to make my point, but those were the arguments made during the hearing, if you get passed all the bullshit.

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site