Sunday, December 25, 2005

Language Learning Feedback

Problem: It is very hard to get the accents right in a foreign language. Although some of the consonants can be tough, vowels seem to be even harder to get right.

Observation: We can easily take a Fourier spectrum of a word and identify the sounds of the various letters in it. That technology is many decades old. Current technology has evolved to allow "training" of voice recognition, so that the computer can identify the individual accents of the user.

Question 1: Can we extend this so that the program can recognize the patterns an individual applies to the underlying sounds? A woman's voice will differ somewhat in timbre from a man's, for example. Can we isolate the effect of that timbre?

Question 2: Can we identify a "pure vowel sound" from a particular accent?

Question 3: Can we then combine the two to predict how an individual ought to pronounce a word in a particular accent or language?

Proposal: Using an interactive sound booth, with microphone and headphones, the student speaks the specified words into the microphone. The computer compares the sound the student has produced with the ideal (or the predicted) sound of the word, and rebroadcasts these back to the student's headphones with some feedback mechanism. I suggest that the feedback be volume: the closer the student reproduces the correct tone, the louder the sound in the headphones. The setup could be: left ear is student, right ear is correct sound.

This way the student gets not just practice speaking the language, but instant (and private) feedback on how well he is getting the accent right.


Post a Comment

<< Home