To economize the frequency spectrum, in telephone communication, the maximum frequency of speech is only 4 kHz. In that frequency spectrum range [0, 4kHz], the human ears tan only hear the tones with enough high spectrum energy, higher than hearing thresholds. Moreover, the tones with lower spectrum energy appearing near (both frequency and time domains) the tones with higher spectrum energy are able to be masked, the human ears cannot hear them. This paper proposes a method to compress telephone speech with almost no loss of sound quality. It is based on the absolute threshold of hearing and distributed to critical bands of the human ears. The telephone speech spectrum is splited into 3 subbands, each subband covers some critical bands of human ear. In each subband, the tones not heard by human ears are omitted. The subbands have lower spectrum energy coded with the fewer bit number and vice versa.