Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Acoustic filtering to best approximate self-voice perception
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Viivi Kanerva1,2, Susanna Girino2, Pavo Orepić3, Defne Abur2; 1University of Turku, Finland, 2University of Groningen, the Netherlands, 3University of Zurich, Switzerland
During speech production, humans perceive their own voice from sound that is transmitted to the cochlea via two different pathways: air conduction (AC) and bone conduction (BC). This renders the perception of self-produced voice (hereafter ‘self-voice’) different from the perception of external voices and other sounds that are conducted only through air. Specifically, BC is assumed to filter out higher frequencies inducing the impression of a lower voice and hence the common experience of one’s voice sounding unfamiliar on a recording. Furthermore, BC conveys a significant portion of self-voice: it has been demonstrated that the AC and BC signals are of approximately equal intensity. Despite the distinct role BC has in self-voice perception, most research in the field has been done solely taking into account the AC component in experimental stimuli and their presentation. Stimuli are typically recorded with conventional AC microphones and played back using AC headphones resulting in a perceptual mismatch between how stimuli sound and how participants hear themselves during natural speech production. Thus, determining the influence of BC contributions to self-voice perception has relevance for production and perception studies of speech and voice. Some studies have used acoustic filters with varying low-frequency weightings to simulate the combination of the AC and BC components in self-voice. Although there was significant variation across participants in the type of acoustic filter they preferred in these studies, most participants found a filtered rather than unfiltered version of self-voice to sound most like themselves. To our knowledge, these types of acoustic filters have not been applied to studies of self-voice discrimination or speech motor control. This is an important aspect to explore since attributing stimuli to being self-generated has been shown to affect speech motor control (i.e., how speech perception is used to guide fluent production) and neural activity is modulated by the perceived similarity between self-voice stimuli and one’s own voice. More generally, self-related stimuli have been associated with prioritization in cognition (e.g., faster perception, enhanced attention, and improved task accuracy). We will present our ongoing study in the Sandbox Series of SNL 2026. We are using five different filters from previous work to simulate the combination of the AC and BC components in self-voice to determine which filter best approximates self-voice perception. Participants listen to unfiltered and filtered versions of their prerecorded speech samples (sustained vowel /a:/ and a sentence) and rate them based on how close their voice in each sample is to their natural speaking voice. Piloting is underway, and data collection (N = 20) and analyses are expected to be completed by SNL. In a later part of the study, we will use the acoustic filter that is most preferred in the first experiment for a self-voice discrimination and speech motor control task. We will test the hypotheses that self-voice discrimination is improved and speech motor control is affected when stimuli sound more like the participants’ voices sound to themselves. The results will yield recommendations for future experimental paradigms of self-voice perception and speech motor control.
Topic Areas: Speech Perception, Methods