Bilingual acoustic voice variation is similarly structured across languages

Abstract

When a bilingual switches languages, do they switch their "voice"? Using a new conversational corpus of speech from early Cantonese-English bilinguals (N = 34), this paper examines the talker-specific acoustic signature of bilingual voices. Following prior work in voice quality variation, 24 filter and source-based acoustic measurements are estimated. The analysis summarizes mean differences for these dimensions, in addition to identifying the underlying structure of each talker's voice across languages with principal components analyses. Canonical redundancy analyses demonstrate that while talkers vary in the degree to which they have the same "voice" across languages, all talkers show strong similarity with themselves.

Bilingual acoustic voice variation: the case of Sorani Kurdish-Persian speakers

“…The method used in this survey is adapted from the works of Johnson et al (2020), , and Lee & Kreiman (2022) on analyzing the voice-quality-related parameters. All analyses were carried out using R version 4.0.4 (R Core Team, 2021)…”

“…The first four formant frequencies (F1, F2, F3, F4): the first four formant frequencies are associated with the transfer function of the vocal tract (Kreiman et al, 2014). The first three formants are commonly employed when discussing the linguistic variations in different languages and the fourth formant is mostly referred to as a speaker-specific parameter (Johnson et al, 2020) H1*-H2*1, H2*-H4*, H4*-H2kHz*, and H2kHz*-H5kHz: all these parameters are associated with the spectral shape of the harmonic source. H1*-H2* denotes the difference between the amplitude of the first and the second harmonics and gauges the harmonic slope which is indicative of the phonation type.…”

“…H2kHz*-H5kHz is the amplitude difference between the closest harmonics to the 2000 Hz and 5000 Hz. This parameter is related to the spectral slope of harmonic independent of F0 (Johnson et al, 2020).…”

“…Their results indicated that bilingual speakers in both language groups exhibited different voice patterns depending on the language. Johnson et al (2020) investigated the degree to which the voice quality of bilingual speakers changes across two languages, namely Cantonese and English. They extracted and measured F0, F1-F4, the corrected versions of harmonic spectral slopes (i.e., H1*-H2*, H2*-H4* respectively), the corrected version of amplitude difference between the fourth harmonic and the harmonic closest to 2000 Hz (i.e., H4*-H2kHz*), the corrected amplitude difference between the harmonics closest to 2000 Hz and 5000 Hz (i.e., H2kHz*-H5kHz*), cepstral peak prominence (CPP), energy, and subharmonics-harmonics amplitude ratio (SHR), using VoiceSauce (Shue et al, 2009).…”

Acoustic voice variation in spontaneous speech

“…However, building mental models of individual voice spaces requires substantial experience with specific talkers. Lacking wide familiarity with how a person's voice varies across changing contexts, it is challenging to recognize that two voice samples come from the same talker ("telling voices together"; Johnson et al, 2020;Lavan et al, 2019b). Consistent with this view, recent evoked potential data (Plant-H ebert et al, 2021) showed that responses to familiar and unfamiliar voices vary with time window as well as familiarity status, implying a two-step process beginning with the evaluation of shared features, and moving on to incorporation of idiosyncratic features as needed for familiar voices.…”