A cross-linguistic study of speech modulation spectra

Area of Science:

• Phonetics and Phonology
• Acoustic Speech Analysis
• Computational Linguistics

Background:

• Languages exhibit systematic variations in sound patterns and grammar, leading to typological classifications like stress-timed vs. syllable-timed and Head-Complement (HC) vs. Complement-Head (CH).
• The acoustic correlates of these broad linguistic categories in speech signals have not been fully understood.

Purpose of the Study:

• To investigate how linguistic typological properties are reflected in the acoustic characteristics of speech across different languages.
• To analyze the amplitude-modulation (AM) and frequency-modulation (FM) spectra of speech to identify language-specific acoustic features.

Main Methods:

• Analysis of AM and FM spectra from 1797 utterances in ten languages.
• Examination of modulation spectra using a perceptually relevant representation based on the modulation index.
• A secondary study on a semi-spontaneous speech corpus to assess the robustness of findings across speakers and speech styles.

Main Results:

• While overall spectral shapes were similar across languages, significant effects of linguistic factors were found in the AM spectra.
• The modulation index effectively distinguished between HC and CH languages (except Turkish).
• The frequency of the modulation index maximum differed between stress-timed and syllable-timed languages.
• These acoustic differences persisted with more speakers but diminished in less constrained semi-spontaneous speech.

Conclusions:

• Broad linguistic categories are indeed reflected in the temporal modulation features of speech.
• Acoustic analysis, particularly of AM spectra and modulation index, can reveal typological differences between languages.
• The manifestation of these differences may be influenced by speaking style and speech constraints.