WHICH TWO ARE THE SAME PHONE?

• We need to extract meaningful information from the signal.

• The speech recognizer tries to model observable statistical variation in the signal.

• These observations are concatenated into feature vectors and directly used in the probability space.