Features

• Forensic voice identification relies on an instrument called the sound spectrograph. The sound spectrograph--available in both analog and digital versions--generates a visual representation of speech patterns known as a voiceprint, according to Richard Saferstein's 2004 textbook "Criminalistics: An Introduction to Forensic Science," 8th edition. The analog version converts voice frequencies into electrical signals and records them on paper using a stylus, notes Saferstein. The digital version works similarly, but without the stylus. Instead, voiceprints display on a computer screen.
  
  

Function

• Identifying a speaker requires forensic audio examiners to conduct both an auditory and visual analysis. They listen to audio evidence to identify distinct or particularly clear portions of recordings, according to a 2009 article in "Signal Processing Magazine." Next, they meet with the suspect and have him speak the same words--with timing as close to the original recording as possible. This may take several attempts. The examiner creates spectrograms to visually compare the voiceprints and look for matches. For example, 20 matching sounds would result in a positive identification, while 15 matches would result in a "probable" identification, according to Saferstein.
  
  

History

• Bell Telephone Laboratories conducted industry research that led to the creation of the sound spectrograph in 1941, says Saferstein. However, as Saferstein notes, it soon found its way into military intelligence when the U.S. military used it to identify German voices transmitted in enemy communications. Lawrence Kersta, an engineer for Bell South, later developed the concept that every human voice has its own voiceprint. Forensic applications followed.
  
  

Misconceptions

• The word "voiceprint" conveys the impression that a spectrogram provides the same reliable identification as a DNA fingerprint. However, forensic voice identification lacks crucial data to make rock-solid conclusions. Even though no two speakers will produce identical voices, this does not mean that forensic audio examiners can identify speakers with certainty, explains "Signal Processing Magazine." Reliable identification must also demonstrate how typical a particular voiceprint is among the general population. Otherwise, a match means very little, since hundreds or thousands of other voices might also produce a similar voiceprint.
  
  

Considerations

• "Signal Processing Magazine" indicates that courts may not admit forensic voice identification into evidence--largely due to the subjective nature of voiceprint analysis and the lack of quantitative studies demonstrating its reliability. Examiners may inadvertently allow bias to creep into results by "loosening" or "tightening" their criteria for positive identifications as a result of experiences with previous cases, cautions a 2005 paper by the Audio Forensics Center.