This study investigates whether fricatives in conversational speech can be accurately identified using random forest modeling based solely on extended high frequencies (EHFs), defined as frequency information above 8 kHz. Previous research suggests that EHFs enhance speech intelligibility and clarity, particularly in noisy environments, and contribute to sound localization and speaker identification in conversational contexts. However, their role in distinguishing specific speech sounds remains unclear. To address this, we analyze high-pass filtered recordings of sociolinguistic interviews in Western Canadian English, focusing on whether EHFs provide sufficient acoustic cues to differentiate fricative consonants in a random forest classification model.
Kharlamov et al. (Tue,) studied this question.