Romance languages such as Italian or Spanish preserve fricative voicing contrasts in word-final position, while their neutralization has been reported for European Portuguese, but the behavior of Romanian fricatives remains understudied. Previous work with Romanian fricatives suggests a pattern of final devoicing but, due to the specific properties of the corpus analyzed, it is unclear if this is limited to the presence of secondary palatalization and/or the result of morphological conditioning. In this study, we apply speech processing tools to investigate the acoustic characteristics of the voicing contrast in fricatives in contemporary spoken Romanian. We examine a corpus of prepared speech from newscasts and semi-spontaneous TV debates (86 speakers) and compare our results to previous findings from a corpus of controlled experimental speech (31 speakers). Our classification tool employs cepstral coefficients and hidden Markov model (HMM)-defined temporal regions to identify the properties of these segments. Our findings conform to typological predictions regarding partial devoicing in coda position, especially at more posterior places, but we find little support for voicing neutralization in Romanian fricatives more generally. Our study thus documents the properties of Romanian fricatives and contributes to our understanding of the dynamics of contrast maintenance in phonological systems.