Academic journal article Perception and Psychophysics

Does Perceptual Learning in Speech Reflect Changes in Phonetic Category Representation or Decision Bias?

Academic journal article Perception and Psychophysics

Does Perceptual Learning in Speech Reflect Changes in Phonetic Category Representation or Decision Bias?

Article excerpt

Recent studies show that perceptual boundaries between phonetic categories are changeable with training (Norris, McQueen, & Cutler, 2003). For example, Kraljic and Samuel (2005) exposed listeners in a lexical decision task to ambiguous /s-∫/ sounds in either s-word contexts (e.g., legacy) or ∫-word contexts (e.g., parachute). In a subsequent /s/-/∫/ categorization test, listeners in the /s/ condition categorized more tokens as /s/ than did those in the /∫/ condition. The effect-termed perceptual learning in speech-is assumed to reflect a change in phonetic category representation. However, the result could be due to a decision bias resulting from the training task. In Experiment 1, we replicated the basic Kraljic and Samuel (2005) experiment and added an AXB discrimination test. In Experiment 2, we used a task that is less likely to induce a decision bias. Results of both experiments and signal detection analyses point to a true change in phonetic representation.

(ProQuest: ... denotes formulae omitted.)

Recent research on speech perception has focused on perceptual learning as a means of solving the problem of variability in speech. According to the perceptual learning account, the idiosyncrasies of a speaker's production of speech sounds are learned and retained rather than mapped onto more abstract representations and discarded. An example of a talker-specific characteristic that might be handled by perceptual learning comes from Newman, douse, and Burnham (2001). They showed that the acoustics of/s/ (as in see) and /∫/ (as in she) vary greatly from speaker to speaker, and that, across speakers, there is extensive overlap between /s/ and /∫/ in the mean frequency of frication noise. However, within a given speaker, the /S/-/∫/ overlap is greatly reduced. With some experience, a listener could learn a speaker's characteristic /s/ and /∫/ frequency ranges and distinguish them in a speaker-specific manner.

A variety of studies have demonstrated the beneficial effect of experience with talker-specific characteristics. Nygaard and colleagues (Nygaard & Pisoni, 1998; Nygaard, Sommers, & Pisoni, 1994) showed improved word identification in noise for voices on which listeners had been trained for several days, and McGarr (1983) found that experienced listeners transcribed the speech of deaf speakers more accurately than did inexperienced listeners. Even very brief exposure to a speaker has been shown to affect perception. In a classic study, Ladefoged and Broadbent provided an early demonstration of rapid adaptation to speaker characteristics (Ladefoged, 1989; Ladefoged & Broadbent, 1957) in which the formant ranges of a precursor sentence determined the vowel perceived in a /b/-V-/t/ test word. More recently, Clarke and Garrett (2004) found an increase in processing efficiency for Spanish- and Chinese-accented speech after less than a minute of exposure.

There is evidence that the benefit of previous experience with a talker is at least in part due to flexibility at the phonetic level of processing. Norris, McQueen, and Cutler (2003) found a shift in Dutch listeners' /s/-/f/ categorization boundary depending on whether an ambiguous IsI-ItI sound replaced /s/s or /f/s in a preceding lexical decision task. If the ambiguous sound occurred in s-final word contexts, then more /s/ responses were given in the categorization test, and vice versa for the f-final word condition. Importantly, there were only 20 samples of the ambiguous sound in the lexical decision task, again demonstrating that perceptual learning in speech can occur quite rapidly. The authors concluded that feedback from lexical representations altered the /s/ and ItI phonetic categories. Subsequent studies have added to our understanding of perceptual learning in speech. The learning is speaker specific in cases of spectral contrast, such as /s/ versus ItI and /s/ versus /∫/ (Eisner & McQueen, 2005; Kraljic & Samuel, 2005, 2007), but generalizes across speakers for nonspectral contrasts, as in the timing contrast in IiI versus /d/ (Kraljic & Samuel, 2006,2007). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.