Speech Recognition with Factorial-HMM Syllabic Acoustic Models