The Accompanying Effect in Speech: How Pitch, Loudness, and Formants Coadjust to Simultaneous Perturbations

Purpose:
This study aimed to further investigate the accompanying effect in speech motor control by examining how pitch, loudness, and formants (first formant [F1]) coadjust at the trial level when auditory feedback is simultaneously perturbed in any two of three acoustic dimensions. Additionally, the study also assessed whether shift direction predictability modulates these cross-dimensional responses.
Method:
Forty-six native Mandarin speakers produced the sustained vowel /ɛ/ for 3 s while auditory feedback was shifted either unpredictably or predictably in single-shift conditions (pitch only, loudness only, and F1 only) and simultaneous-shift conditions (pitch–loudness [PL], pitch–F1 [PF], and loudness–F1 [LF]). We quantified response proportions (opposing vs. following) and behavioral features (onset time, peak time, and relative gain) for each acoustic dimension at the trial level. A 13.5% threshold was used to distinguish frequent from infrequent response types based on predicted distributions from Bayesian Poisson regression modeling. Response latency and relative gain were analyzed using linear mixed-effects modeling.
Results:
The high percentage of cross-dimensional responses at the trial level (95% for the single-shift conditions and 96% for the simultaneous-shift conditions) confirmed a robust accompanying effect. Within-domain opposing responses remained the most prominent response type. For the unperturbed domain, speakers frequently adjusted in systematic ways (e.g., formant raising with PL shifts and loudness reduction with PF shifts). Pitch responses were the fastest and largest, F1 responses were the slowest and smallest, and loudness responses exhibited an intermediate profile (fastest onset, slowest peak, moderate gain). Shift direction predictability increased opposing responses only in the PL and LF conditions.
Conclusions:
These findings demonstrate that the speech motor systems for regulating pitch, loudness, and formants are coupled rather than entirely independent. Our results underscore the need to extend existing speech models to account for cross-dimensional coupling and the heightened susceptibility to pitch perturbations.