LEARNING TO PRONOUNCE VOWEL SOUNDS IN A FOREIGN-LANGUAGE USING ACOUSTIC MEASUREMENTS OF THE VOCAL-TRACT AS FEEDBACK IN REAL-TIME

An acoustic impedance spectrometer was used to measure the frequencies R1 and R2 of the first two resonances of the Vocal tract. The measure ment was made just outside the mouth, in parallel with the free field, using a new technique that provides precise information about the aco ustic response of the vocal tract in real time. Values measured for na tive speakers for a particular vowel were used as target parameters fo r subjects who used a visual display of an impedance spectrum of their own vocal tracts as realtime feedback to realize the vocal tract conf iguration required to pronounce the target vowel. We report the values (R1,R2) for eleven non-nasalized vowels of French. These values are s imilar to the formant frequencies measured previously for these vowels , and their relative positions in the (R2,R1) plane are similar to tho se of the same vowels in the (F2,F1) formant plane. The confusion and correct identification of these vowels are shown to be strongly relate d to their separation in the (R2,R1) plane. We report the results of a ttempts to imitate six of these vowels by monolingual anglophone subje cts. One group used a traditional method of learning pronunciation: th ey heard the vowel sounds and then attempted to imitate them. Another group also heard the sounds, but were assisted by the vocal tract feed back described above when imitating the target sounds. The acoustic pr operties and recognizability of the vowels were significantly superior when the subjects used vocal-tract feedback.