Improving speaker de-identification with functional data analysis of f0 trajectories

Due to a constantly increasing amount of speech data that is stored in different types databases, voice privacy has become major concern. To respond such concern, researchers have developed various methods for speaker de-identification. The state-of-the-art solutions utilize deep learning which can be effective but might unavailable or impractical apply for, example, under-resourced languages. Formant modification simpler, yet method de-identification requires no training data. Still, remaining intonational patterns formant-anonymized may contain speaker-dependent cues. This study introduces novel method, which, addition simple formant shifts, manipulates f0 trajectories based on functional analysis. proposed will conceal plausibly identifying pitch characteristics phonetically controllable manner and improve formant-based up 25%.