PURPOSE: To emphasize a spectral peak by analog signals and to compress a frequency band by dividing signals whose feature quantity is the spectral peak to the plural frequency bands, limiting amplitude for the respective frequency bands and extracting frequency components.

CONSTITUTION: 200-3400Hz are divided into seven frequency bands by band-pass filters 3A-3G and input speech signals are shaped into rectangular wave signals in amplitude limiters 4A-4G. Then, by the sub band rectangular waves of 40-680Hz frequency divided by frequency dividers 6A-6G, carrier waves are amplitude-modulated in modulators 7A-7G and outputted. Thus, output is performed as frequency divided sub band signals whose amplitude is the same as sub band signals and frequency is 1/5. Then, the signal components of respective frequency divided sub bands are extracted, synthesized and outputted by the band-pass filters 8A-8G. Thus, the speech signals are compressed to the narrow frequency band of 40-680Hz and the spectral peak of the feature quantity is faithfully reproduced.