Acoustic variation is central to the study of speaker characterization. In this respect, specific phonemic classes such as vowels have been particularly studied, compared to fricatives. Fricatives exhibit important aperiodic energy, which can extend over a high-frequency range beyond that conventionally considered in phonetic analyses, often limited up to 12 kHz. We adopt here an extended frequency range up to 20.05 kHz to study a corpus of 15 812 fricatives produced by 59 speakers in Russian, a language offering a rich inventory of fricatives. We extracted two sets of parameters: the first is composed of 11 parameters derived from the frequency spectrum and duration (acoustic set) while the second is composed of 13 mel frequency cepstral coefficients (MFCCs). As a first step, we implemented machine learning methods to evaluate the potential of each set to predict gender and speaker identity. We show that gender can be predicted with a good performance by the acoustic set and even more so by MFCCs (accuracy of 0.72 and 0.88, respectively). MFCCs also predict individuals to some extent (accuracy = 0.64) unlike the acoustic set. In a second step, we provide a detailed analysis of the observed intra- and inter-speaker acoustic variation.

This paper shows that machine learning techniques are very successful at classifying the Russian voiceless nonpalatalized fricatives [f], [s], and [S] using a small set of acoustic cues. From a data sample of 6320 tokens of read sentences produced by 40 participants, temporal and spectral measurements are extracted from the full sound, the noise duration, and the middle 30 ms windows. Furthermore, 13 mel-frequency cepstral coefficients (MFCCs) are computed from the middle 30 ms window. Classifiers based on single decision trees, random forests, support vector machines, and neural networks are trained and tested to distinguish between these three fricatives. The results demonstrate that, first, the three acoustic cue extraction techniques are similar in terms of classification accuracy (93% and 99%) but that the spectral measurements extracted from the full frication noise duration result in slightly better accuracy. Second, the center of gravity and the spectral spread are sufficient for the classification of [f], [s], and [S] irrespective of contextual and speaker variation. Third, MFCCs show a marginally higher predictive power over spectral cues (<2%). This suggests that both sets of measures provide sufficient information for the classification of these fricatives and their choice depends on the particular research question or application.

No abstract