Ranking the Pitches of Concurrent Vowels

In this study we examined the interaction of fundamental frequency (Fo) differences and binaural separation cues on the identification accuracy and judgments of the relative pitches of concurrent vowel pairs. On each pitch ranking trial listeners were told which vowels were presented and were asked simply to indicate which vowel was on the higher pitch. The first vowel had an FO near 140 Hz and the second vowel was either 1,2,4, or 8 semitones lower or higher than the ftrst. When the Fo difference was small the ability to rank the pitches of the two vowels was close to chance level, and dichotic presentation provided little advantage over monaural presentation. Performance improved gradually with increasing F. difference, reaching a maximum of around 80% correct at +8 semitones. For vowel identification, dichotic presentation resulted in consistently higher vowel identification accuracy; but dichotic presentation resulted in improved pitch ranking only for positive Fo differences. Differences between pitch ranking and vowel identification in their sensitivity to Fo differences and binaural cues places important constraints on models of concurrent vowel segregation. INTRODUCTION AND EXPERIMENT When two vowels are presented simultaneously, their identification is improved if they are played on different fimdamental frequencies (AFo’s) (8, 2, 3, 7). Similarly, when sentences are processed using a speech vocoder, synthesized with a monotone pitch and played back in pairs, their intelligibility improves if they have different F,-,‘s (5). Bird and Darwin (1997) recently compared the results of experiments with such “double vowels” and “double sentences” and concluded that the patterns of improvement with AFO are quite different. Double vowels show most of their improvement over the first semitone of Fo separation (where double sentences show less improvement) while double sentences continue to show an advantage as the AFO increases beyond the separation at which double vowel identification reaches a plateau. They proposed that double vowels improve because of waveform interactions or beats that occur primarily with small AFo’s (7, 3) while double sentences involve a process of acrosschannel grouping on the basis of Fo. They considered the possibility that larger FO differences are required for the perceptual segregation of double sentences because the individual partials of the two voices overlap extensively in the low-frequency region, making it difYicult to perceive the pitches of the two voices (6, 1). In the present study we examined the ability of listeners to rank the pitches of double vowels, a task which requires judgments of vowel identification as well as pitch. We then compared listeners’ pitch ranking judgments with vowel identification accuracy (both vowels correct) as a function of Al& In addition, we investigated the interaction between AFO and binaural cues provided by dichotic presentation (one vowel to each ear) compared to monaural presentation (both vowels to the same ear). The stimuli were 200-ms double vowels with formant patterns used in previous studies of double-vowel identification (3) and pitch matching (1). The individual vowels were synthesized versions of American English li/, JaJ, JuJ, JzJ, and JzJ, generated with a cascade formant synthesizer. Double vowels were constructed by summing pairs of single vowels. In each pair, one vowel had an F,, of 140 Hz with a random deviation of up to f2 semitones, in % semitone steps. The other F0 differed by f8, ~t4, ti, or fl semitones. All pairwise combinations of the five vowels were used, except that identical pairs (such as /i/ + Ji/) were excluded, for a total of 160 stimuli (20 vowel pairs x 8 F,‘s). Stimuli were presented over headphones and responses were made using a computer mouse with a button box displayed on the computer screen. In monaural conditions, both vowels were presented to the left car; in dichotic conditions, one vowel was presented to each ear. Four separate groups of subjects were randomly assigned to either the monaural or dichotic conditions and either pitch ranking or vowel identification tasks (60 subjects total). Prior to the main experiment each subject completed a 40-item practice session. Each subject then completed 6 trials in the main experiment. In the pitch ranking task, the identities of the two vowels were given by text labels on two button boxes and subjects were asked to select the vowel on the higher pitch. In the vowel identification task subjects were asked to ident@ the two vowels that were present.