Prior to the current funding contract for phase one of this project, a data base had already been developed consisting of about 2,000 recordings of 435 speakers variously producing 27 classes of utterances. The vectors that have been tested were selected on the basis of (1) high probability of discriminating among speakers, (2) enhanced utility when combined with other factors, (3) resistance to distortion, (4) availability, (5) convenience in modification, and (6) compatability with computer processing. The vectors currently being tested are the speaker fundamental frequency vector, the long-term speech spectra vector, the vowel formant tracking vector, the temporal analysis vector, and the vocal intensity vector. Additional vectors are being subjected to pilot-level evaluations. These include vocal jitter, voice shimmer, phoneme analysis, and vowel/consonant ratios. Reasonable identification power (excellent in some instances) was found for all vectors, even in the face of a severely structured forensic evaluation. Plans for the other phases of the experiment are presented.
Downloads
Similar Publications
- Just Science Podcast: Just All Hands On DECK for Drug Endangered Youth
- The Cross-Reactivity of the Cannabinoid Analogs (delta-8-THC, delta-10-THC and CBD) and their metabolites in Urine of Six Commercially Available Homogeneous Imm
- Is the Gender Gap in Overdose Deaths (Still) Decreasing? An Examination of Opioid Deaths in Delaware, 2013–2017