To reduce the amount of calculation and the capacity of a ROM, and to obtain good synthesized sounds in a voice decoding device and a voice coding device used in efficient decoding and coding operations of voice information.

In the voice decoding device, the generation information on the sound source vector constituted of a small number of pulses is stored in a sound source information storage section 1. Moreover, any one of the sequence of numbers i.e., a random numeral column, the sequence of numbers obtained by a learning or the sequence of numbers obtained by knowledge, is stored in a spread vector storage section 2. Spread sound source vectors are then generated by convoluting the sound source information of the section 1 and the spread vectors of the section 2 by a convoluting section 3. The voice coding device is constituted to generate a new impulse response by convoluting the spread vectors and the impulse response from a digital filter in a convolution section.