Amplitude envelope modulations across languages reflect prosody

Abstract

The speech signal has been shown to contain a fine structure that consists of the fast changing spectral content (e.g., formant transitions, voicing, spectral energy distributions), together with amplitude modulations of the envelope with different timescales. These different modulation frequencies have been associated with linguistic units of different sizes, and neuronal oscillations seem to track this linguistic structure. As the amplitude envelope mostly captures suprasegmental information, the different modulation frequencies are natural candidates to convey prosodic information. In this paper we put these assumptions to the test by comparing effects of sentence length and language, focusing on languages with distinct prosodic profiles: Brazilian Portuguese (syllable-timed), European Portuguese (mixed rhythm, with syllable-timed and stress-timed properties), and German (stress-timed). There are further differences regarding the roles of the syllable, the foot, the prosodic word and the intonation phrase. We analyzed wideband amplitude envelopes using general additive mixed models and show that German differs from Brazilian Portuguese and European Portuguese in the delta (1-2Hz) and theta bands (6-8Hz). European and Brazilian Portuguese also differ, but only in the delta band (1-2Hz). The language differences in amplitude modulation are discussed in terms of speech rhythm and differences in prosodic structure across languages.