Date: Tue, 26 Sep 1995 15:50:03 BST


Subject: American English Dialect Samples

Dear Linguist Subscribers,

I am posting this for a colleague here at Daimler-Benz research who is

interested in quantifying the performance of a speech recognition system for


We are interested in getting a representative sample of American English

dialects to train the system and wonder how to do this. Ideally, we would like

to record speakers at less than 10 locations in the US and accumulate between

50 and 100 samples. The speakers will all read from the same list of target


Our question is: at what locations should we do this ? Do you feel that a

representative sampling is possible with this number of speakers ? Any other

hints ?

Please reply directly to my colleague: reh[AT SYMBOL GOES HERE]

We will post a summary to the list.

Many thanks for your help,

A. Kuehnle