Let me say that after reading the article, I am not buying the conclusion that it is not Zimmerman's voice.
First let me provide some of my background experience, prior to focusing on financial software development, I spend the first 15 years of my career involved in telecom and voice software. I wrote code for voice mail systems, voice companding, dsp code, voice recognition, text to speech, T1/E1 digital links, codec, analog echo cancellation and VoIP packetization.
The first problem I have with the conclusion is that the police 911 system probably uses voice compression when recording voice calls digitally which causes a loss of quality to minimize disk storage space requirements. In other words both 911 calls (Zimmerman's and the screaming) have quality issues that make them difficult to use for voice biometric identification.
The second issue is that the call with the screaming probably is from a wireless home handset which uses some type of compression between the handset and the base causing quality issues and distortion. Similarly Zimmerman' cell call has the same problem but in the case of cell phones it is easier to correct for since a common known algorithm is used for cell phone communication.
The third problem is that there is no sample of Trayvon Martin's voice used for comparison in an attempt to demonstrate a match. This is a critical issue.
The fourth problem is that a rating of 46% with the Easy Voice Biometrics software is "inconclusive" not negative. Even if the voice definitely was Zimmerman's it is not surprising that a 46% score is achieved after all the compression that occurred at the 911 call center, wireless handset, and cell. Unless the items are corrected for then the score is going to be low. And to correct for them, the algorithm that is being used for the voice storage and handset communication must be known, simply trying to judge quality based on the signal-to-noise ratio is not sufficient.
As a follow-up note, I will note that in the Davalloo case that was referenced, the criminal called from a pay phone in 2002 that did not use voice compression (only u-law companding). The 911 call center at the time used a Nortel or Lucent analog tape recorder with no digital voice compression. This made the identification of Davalloo much more straight forward. I question the statement that âthe audio from Zimmerman's call is much better quality than the 911 call in the Davalloo caseâ.
My conclusion, his result that "you can say with reasonable scientific certainty that it's not Zimmerman" does not hold true. I think that Owen Forensic Services LLC is simply trying to get publicity to drum up more business.
// Also I will note that Trayvon Martin's father said in an interview that the screaming is not his son's voice.
