Perceptual Equivalence of Two Kinds of Ambiguous Speech Stimuli

Stimuli from two synthetic Ida/-/gal continua were presented in a speeded labeling task. One continuum was generated by parameter interpolation; the other, by adding the waveforms of the endpoint stimuli in varying proportions. Both continua showed an increase in latencies at the category boundary, suggesting that the two procedures yield equally ambiguous stimuli. Ambiguous stimuli play a central role in speech perception research. By virtue of their perceptual instability, they serve as indicators of a large variety of laboratory phenomena, including categorical perception, selective adaptation, phonetic trading relations, and all sorts of context effects. Traditionally, ambiguous stimuli have been constructed with the aid of speech synthesizers: Two unambiguous stimuli from different phonetic categories are selected, and a number of steps are interpolated between their parameter values, leading to a continuum that includes some ambiguous stimuli in the region of the phonetic category boundary. Until recently, this was the only method available. However, a new technique was applied in a recent doctoral thesis by Stevenson (1979). Instead of interpolating parameter values between two endpoint stimuli, he added the digitized waveforms of the endpoint stimuli in various proportions, increasing the amplitude of one component waveform while decreasing that of the other and so producing a continuum. In fact, he was able to construct such continua from carefully aligned natural utterances of Iba/, Ida/, and Iga/; but the technique can, of course, be used with synthetic speech as well. Electronically mixed synthetic stimuli have been used previously, primar~ By to compare their perception with that of the same component stimuli presented dichotically (Halwes, 1969; Porter & Whittaker, 1980; Repp, 1976, 1980). However, Stevenson (1979) was apparently the first to construct whole stimulus continua that way. His technique is interesting, especially because it can be used with natural speech. However, are there any important perceptual differences between an ambiguous stimulus created by superimposing two unambiguous stimuli and one characterized by a single set of intermediate parameters? Stevenson used his stimuli in a variety of standard experimental tasks, including categorical perception, selective adaptation, and dichotic listening, and obtained results very similar to those found with traditional stimulus continua, although he never performed any direct comparison.1 *TO be published in the Bulletin of the Psychonomic Society. Acknowledgment: This research was supported by NICHD Grant HD01994 and BRS Grant RR05596 to Haskins Laboratories. I am grateful to Dale Stevenson for bringing his work to my attention and for commenting on an earlier version of this paper. [HASKINS LABORATORIES: Status Report on Speech Research SR~66 (1981)] 79 80 The present study explored one way in which the two types of ambiguous speech stimuli might differ in perception. When presented with an ambiguous stimulus of the traditional kind, which has acoustic properties that are truly intermediate, listeners experience uncertainty that increases the time needed to assign the stimulus to one of two categories (Studdert-Kennedy, Liberman, & Stevens, 1963; Pisoni & Tash, 1974). However, when listening to a stimulus from a Stevenson continuum, which contains two unambiguous sets of cues superimposed, there might be no uncertainty on a given trial; rather, perception might go wi th one or the other set of unambiguous cues on a probabilistic basis. The present. study tested this hypothesis by examining whether the characteristic peak in identification latencies at the category boundary of traditional speech continua (Studdert-Kennedy et al •• 1963; Pisoni & Tash. 1974) is present to the same extent on a continul,lffi of electronically mixed stimuli.