Canonically, the “r” in Turkish is produced as an alveolar tap; however, word-finally, this tap has been described as devoiced and fricated. Previous work found that mel frequency cepstral coefficients (MFCCs) show spectral differences between canonical and fricated taps in Turkish. Furthermore, work on other languages found MFCCs to be better at discriminating places of articulation for fricatives than spectral moments. Although MFCCs are useful in speech recognition and acoustic research, their phonetic interpretations are still unclear. Current work has been done to interpret MFCCs for vowels in American English [Iskarous, K. however, more work is needed to interpret MFCCs for consonants. MFCCs can be interpreted as providing information about the concentrations of energy in the spectrum, which is useful for identifying fricatives. Using the Turkish dataset from the Mozilla Common Voice corpus, this study takes a closer look at the significant MFCCs for distinguishing fricated and non-fricated rhotics in Turkish. The results will be discussed with respect to how MFCCs might be phonetically interpreted for frication, with a specific focus on the Turkish rhotic.
Miklas et al. (Fri,) studied this question.