Gender identification from high-pass filtered vowel segments: The use of high-frequency energy

Abstract

The purpose of this study was to examine the use of high-frequency information for making gender identity judgments from high-pass filtered vowel segments produced by adult speakers. Specifically, the effect of removing lowerfrequency spectral detail (i.e., F3 and below) from vowel segments via high-pass filtering was evaluated. Thirty listeners (ages 18-35) with normal hearing participated in the experiment. A within-subjects design was used to measure gender identification for six 250-ms vowel segments (/ae/, /ɪ /, /ɝ/, /ʌ/, / ɔ/, and /u/), produced by ten male and ten female speakers. The results of this experiment demonstrated that despite the removal of low-frequency spectral detail, the listeners were accurate in identifying speaker gender from the vowel segments, and did so with performance significantly above chance. The removal of low-frequency spectral detail reduced gender identification by approximately 16 % relative to unfiltered vowel segments. Classification results using linear discriminant function analyses followed the perceptual data, using spectral and temporal representations derived from the high-pass filtered segments. Cumulatively, these findings indicate that normalhearing listeners are able to make accurate perceptual judgments regarding speaker gender from vowel segments with low-frequency spectral detail removed via high-pass filtering. Therefore, it is reasonable to suggest the presence of perceptual cues related to gender identity in the high-frequency region of naturally produced vowel signals. Implications of these findings and possible mechanisms for performing the gender identification task from high-pass filtered stimuli are discussed.

Classification of indexical and segmental features of human speech using low- and high-frequency energy

“…These results are expected and coincide with behavioral findings reported by Monson et al (2014), who found approximately 92% accuracy among listeners with normal hearing when judging speaker gender from sentences high-pass filtered near 5.7 kHz. Donai and Lass (2015) reported correct gender identification near 85% for 250-ms vowel segments highpass filtered at 3.5 kHz. Taken together, the previous behavioral results and current classification findings support information related to talker sex present in the highfrequency portion of the speech spectrum.…”

“…This recent interest extends beyond auditory access to spectral energy present in phonemes produced with frication noise, such as /s/ and /z/, as in previous years (e.g., Stelmachowicz et al, 2001Stelmachowicz et al, , 2002. Segmental information related to phoneme and sentence recognition (e.g., Donai and Paschall, 2015;Vitela et al, 2015) as well as indexical features of the talker, such as sex (e.g., Monson et al, 2014;Donai and Lass, 2015;Donai and Halbritter, 2017) and talker identity (Schwartz et al, 2018), have also been found in the high-frequency region of the speech signal. Some studies limited auditory access only to high-frequency energy (HFE) above 3-5 kHz (e.g., Donai and Paschall, 2015;Donai and Halbritter, 2017;Monson et al, 2014;Vitela et al, 2015), while others varied the spectral bandwidth of the stimuli to purposely include or exclude HFE above 8 kHz (e.g., Schwartz et al, 2018, Flaherty et al, 2021Polspoel et al, 2022).…”

Automated Classification of Vowel Category and Speaker Type in the High-Frequency Spectrum

“…Results showed that listeners were able to use information from the high-frequency spectrum to identify speaker gender with accuracy significantly above chance (approximating 90% accuracy). In addition, Donai and Lass 6 recently reported gender recognition abilities among normal-hearing listeners using 250 ms vowel segments high-pass filtered at 3.5 kHz (above F3). Results indicated that normal-hearing listeners were able to identify speaker gender from signals comprised solely of high-frequency energy with greater than 80% accuracy.…”