A corpus of noise-induced word misperceptions for Spanish

Abstract

Word misperceptions are valuable in designing and evaluating detailed computational models of speech perception, especially when a number of listeners agree on the misperceived word. The current paper describes the elicitation of a corpus of Spanish word misperceptions induced by different types of noise. Stimuli were presented using an adaptive procedure designed to promote the rapid discovery of misperceptions. The final corpus contains 3235 misperceptions along with speech and masker waveforms, permitting further experimental and modeling studies into the origin of each misperception. The corpus is available online as an open resource.