Forensic Speaker Verification Using Ordinary Least Squares

Abstract

In Brazil, the recognition of speakers for forensic purposes still relies on a subjectivity-based decision-making process through a results analysis of untrustworthy techniques. Owing to the lack of a voice database, speaker verification is currently applied to samples specifically collected for confrontation. However, speaker comparative analysis via contested discourse requires the collection of an excessive amount of voice samples for a series of individuals. Further, the recognition system must inform who is the most compatible with the contested voice from pre-selected individuals. Accordingly, this paper proposes using a combination of linear predictive coding (LPC) and ordinary least squares (OLS) as a speaker verification tool for forensic analysis. The proposed recognition technique establishes confidence and similarity upon which to base forensic reports, indicating verification of the speaker of the contested discourse. Therefore, in this paper, an accurate, quick, alternative method to help verify the speaker is contributed. After running seven different tests, this study preliminarily achieved a hit rate of 100% considering a limited dataset (Brazilian Portuguese). Furthermore, the developed method extracts a larger number of formants, which are indispensable for statistical comparisons via OLS. The proposed framework is robust at certain levels of noise, for sentences with the suppression of word changes, and with different quality or even meaningful audio time differences.

Linguistic document research technologies (forensic and procedural aspects)

“…Voice samples obtained through such specific recommendations are usually very suitable for comparison purposes, but there is a universal standard for the number of words required for identification. It can range from 10 to 20 words for various situations (Machado et al, 2019). Using spectrographic voice identification technology, it can be stated that the variability within the technical means that records and reproduces oral speech differs from the variability between different technical means.…”

Forensic Speaker Verification: Case Report at Brazilian Justice

“…Voiceover identification through voice audio files has, for base, the comparison between the disputed audio and an audio file containing the same sentence recorded on the disputed audio file, which is called reference audio. The methodology used at the implementation of OLS algorithm demands some complementary steps, be it for analysis or final verification to obtain results, just like presented on Figure 1 [16]. Thus, audio files are initially windowed and transformed for the frequency domain for Linear Predictive Coding (LPC) parameters generation.…”

“…Audio processing was conducted using an application developed in Matlab 2018b [16]. First, data were loaded by a guest user interface (GUI), where the user could select the desired audio file.…”

“…Likewise, speaker recognition in Brazil still presents decision-making based on the subjective analysis of results using unreliable techniques [15]. To overcome that, a methodology based on analyzing formants via OLS algorithm was proposed in [16]. The authors stated that the obtained results are legitimate for Brazilian Portuguese and these strongly reinforce the potentiality of the methodology employed since the several voic-es dataset is in English, which could provide distance from Brazilian Portuguese.…”

“…Based on all the above, this report analyzes the performance of a technique proposed on forensic speaker verification based on the OLS algorithm [16], for assertive efficiency as well as for response time. Five cases, which have already prosecuted and sentenced by the Mato Grosso court in Brazil, were employed to evaluate the method.…”