Toward Spontaneous Speech Recognition and Understanding