A dictation system is disclosed comprising a hand held dictation device (1) for storing a speech signal in memory means (15,20), the device comprising data compression means (30) for data compressing the speech signal into a data compressed speech signal and storing means for storing the data compressed speech signal in the memory means. The data compression means (30) are adapted to carry out a data compression step on the speech signal in one of at least two different data compression modes, the at least two different data compression modes resulting in different data compression ratios when applied to the same speech signal, the said at least two different data compression modes being selectable by a user. The data compression means (30) are further adapted to create data files (Bi) comprising portions of the data compressed speech signal, each of the data files comprising a header portion (HDR), the data compression means being also adapted to generate an identifier signal identifying the data compression mode selected and being adapted to store said identifier signal in said header portion.