Sub-band parametric cepstral distance measurement of voiceless alveolar fricative segments as a tool for identifying speaker-characteristic information robust to emotional variation

Existing methods of identifying speaker-specific acoustic information in Forensic Voice Comparison (FVC), such as formant-based methods, suffer in performance when dealing with variation in speaker emotion. This is especially the case when it comes to the extremities of emotionally conditioned speech variation which often appear in FVC casework. Therefore, it is worthwhile to ask the question whether there exist particular aspects of speech data which are indicative of speaker-specific characteristics, but which remain robust to variation in emotional state. We present our investigation into the potential utility of the segment /s/ for this purpose, using the Parametric Cepstral Distance (PCD) method proposed by Clermont and Mokhtari (1994). We find that the cepstrum seems to contain speaker-specific information around 2–3kHz, a range which also demonstrates strikingly little emotionally-conditioned variation. We therefore suggest that the spectral characteristics of /s/ could be useful for FVC, where analysts are often required to compare emotionally varied recordings. The segment /i:/ is also used in order to ensure the reliability of the emotion coding groups established.