Intra- and inter-speaker variation in eight Russian fricativesa)

Area of Science:

• Phonetics and Speech Science
• Acoustic Analysis
• Speaker Characterization

Background:

• Vowels are well-studied in speaker characterization, while fricatives receive less attention.
• Fricatives possess aperiodic energy extending beyond conventional phonetic analysis frequencies (12 kHz).
• Russian language offers a rich inventory of fricative sounds for study.

Purpose of the Study:

• To investigate acoustic variation in fricatives using an extended frequency range (up to 20.05 kHz).
• To evaluate the effectiveness of acoustic and Mel Frequency Cepstral Coefficients (MFCCs) in predicting speaker gender and identity.
• To analyze intra- and inter-speaker acoustic variations in fricatives.

Main Methods:

• Utilized a corpus of 15,812 Russian fricatives from 59 speakers.
• Extracted two parameter sets: acoustic features (frequency spectrum, duration) and MFCCs.
• Applied machine learning models to predict gender and speaker identity.

Main Results:

• Gender prediction accuracy: 0.72 (acoustic set) and 0.88 (MFCCs).
• Speaker identity prediction accuracy: 0.64 (MFCCs), while the acoustic set showed limited success.
• Detailed analysis of intra- and inter-speaker acoustic variation was performed.

Conclusions:

• MFCCs demonstrate superior performance for predicting gender and speaker identity from fricatives compared to traditional acoustic features.
• The extended frequency range provides valuable insights into fricative acoustics for speaker characterization.
• Further analysis of acoustic variation can enhance speaker recognition technologies.