The effects of anger on automated long-term-spectra based speaker-identification

TL;DR

This study reveals that anger significantly distorts acoustic signals in long-term spectra-based speaker identification, reducing accuracy by about one-third, which suggests the need for caution in forensic applications.

Contribution

It demonstrates that emotional states, specifically anger, can impair the reliability of long-term spectra-based speaker identification methods.

Findings

Anger causes a 33% decrease in identification accuracy.

Moderate anger can lead to misidentification as a different speaker.

Long-term spectra analysis is sensitive to emotional distortions.

Abstract

Forensic speaker identification has traditionally considered approaches based on long term spectra analysis as especially robust, given that they work well for short recordings, are not sensitive to changes in the intensity of the sample, and continue to function in the presence of noise and limited passband. We find, however, that anger induces a significant distortion of the acoustic signal for long term spectra analysis purposes. Even moderate anger offsets speaker identification results by 33% in the direction of a different speaker altogether. Thus, caution should be exercised when applying this tool.