abstract Andreas Lind

Andreas Lind (Lund University, Sweden)

Real-time Speech Exchange

Most models of speech production are comparator models, which hold that speech starts with a clear preverbal conception of what to say that anchors our sense of agency over speech and is used as a benchmark for self-monitoring. According to an alternative, inferential, model, speakers often have no such detailed preview of what they are about to say, but instead use feedback from their own utterances to help specify the meaning of what they just said. During my time as a PhD-student, my colleagues and I developed a new technique for feedback manipulation during speech which we call Real-time Speech Exchange (RSE), to try to investigate this issue. Participants wear specially designed sound isolating headsets. RSE allows us to surreptitiously record words or phrases that they utter, and then to play this recording back to them later in the test at the exact same time as they utter another word. As we simultaneously block out the feedback of what they are actually saying, we create situations where participants say one thing, but receive real-time, timing matched auditory feedback suggesting they are saying something else. In one experiment, we used RSE to show that speakers often do not detect manipulations, and that they often believe themselves to have said the inserted word, rather than the word they actually said. This we take to indicate that there is a strong inferential component in the agency processing of speech, and that auditory feedback one’s own voice can act as a pathway for high-level, semantic self-monitoring, potentially overriding other feedback loops when we specify to ourselves the meaning of what we say. In my talk, I will discuss the studies we have done so far, bring up some criticism that have been directed at them, and point towards some of the things we will do in the future.