Robustness of Fundamental Frequency under Mobile Phone Codec for Estimating Likelihood Ratios in Forensic Voice Comparison

Abstract

This preliminary study will investigate the robustness of fundamental frequency (F0) as a forensic voice comparison (FVC) feature in estimating likelihood ratios (LRs) with mobile phone speech. F0 is one of the most commonly used features in FVC with some studies indicating its robustness against environmental noise and communication transmission [1], while some other studies point out that F0 is prone to various intrinsic and extrinsic factors, including emotions, intoxication, stress, etc. [2]. Investigations from [3] reported that a mobile phone network affected the distribution of F0 values for some speakers but not others. To the best of our knowledge, all relevant previous studies investigated the influence of mobile phone channels on the F0 acoustics, and there has not been any studies in which the impact of a mobile channel on the magnitude of LRs with F0-based features was investigated. The current study concerns the adaptive multi rate (AMR) codec, which is one of the most widely used codecs in the global system for mobile communications (GSM). This study is particularly interested in one of the key features of the mobile phone technologies; namely dynamic rate coding (DRC) which changes the source coding bit rate on a frame-by-frame basis according to changing channel quality within a GSM network [4]. Thus, this study looks into how the different source coding bit rates will impact on 1) the overall performance of the LR-based system with F0 features, and also on 2) the magnitude of the LRs that are estimated by the same system.