NCJ Number
219473
Journal
Journal of Forensic Identification Volume: 57 Issue: 4 Dated: July/August 2007 Pages: 582-598
Date Published
July 2007
Length
17 pages
Annotation
This study analyzed test recordings prepared on 10 small solid-state audio records to determine the optimal playback and the effects that these compressed digital audio recordings had on forensic analyses.
Abstract
Results indicated that none of the playback methods were consistently better than the others when comparing frequency responses and intermodulation distortion. Comparisons of the signal-to-noise ratios of the headphone outputs, the converted WAV files, and the direct playback indicated that in most cases, the digital outputs were better. Other findings revealed that most recorders had a sampling frequency variance of less than 0.02 percent. Vocal resonances were lost at higher frequencies and some added distortion artifacts. A number of recommendations are made relating to the authenticity, enhancement, signal analysis, voice comparisons, and related forensic examinations. These recommendations include the caution that one of the units recorded higher-frequency sounds, which could affect gunshot analysis. The 10 test recorders were all small solid-state devices with an integrated flash memory chip that ranged in capacity from 4 to 256 megabytes. They were purchased in the United States and ranged in cost from $35 to $375. The test recordings had four components: (1) a 1-minute recording of a 500.00 hertz (Hz) discrete sine wave tone; (2) a 1-minute recording of equal-level 500.00 and 900.00 Hz discrete sine wave tones; (3) a 1-minute recording of “white” noise; and (4) 42 voice samples taken from the Texas Instruments/Massachusetts Institute of Technology (TIMIT) database consisting of 21 different speakers speaking 2 phonetically rich sentences. Test recordings were placed in a single WAV file using audio-editing software and each sample was separated by 5 seconds of silence. All speech samples were resampled to 44,100 Hz to match the nonvoice signals. All data were then normalized to the same maximum amplitude. Test recordings were then outputted from the recorders using up to three methods: (1) analog headphone output; (2) USB output; and (3) direct playback of native files in the manufacturer’s software. Tables, figure, references