•     The signal is converted to a sequence of    feature vectors based on spectral and    temporal measurements.
				•     Acoustic models represent sub-word    units, such as phonemes, as a finite-    state machine in which states model    spectral structure and transitions    model temporal structure.
				•     The language model predicts the next       set of words, and controls which models    are hypothesized.
				•     Search is crucial to the system, since       many combinations of words must be      investigated to find the most probable       word sequence.