We describe a recognition experiment and two analytic experiments on a database of strongly Hispanic-accented English. We show the crucial importance of training on the Hispanicaccented data for acoustic model performance, and describe the tendency of Spanish-accented speakers to use longer, and presumably less-reduced, schwa vowels than native-English speakers.

We studied the effect of MLLR adaptation with Spanishaccented English to understand the strengths and weaknesses of MLLR with unseen foreign accents. We trained a global MLLR transform on 10 adaptation sentences per speaker, giving a 3.4 % absolute decrease in phone error rate. We then studied the pattern of improvements across phones and phone classes. Phones that improved the least tended to be those that do not exist in Spanish. Results suggest the poorer performance is related to increased insertion and substituter rates during the adaptation phase, as well as greater acoustic variability. Index Terms: speech recognition, adaptation, MLLR, foreign accent, Spanish, error analysis