A speech encoder includes an LPC synthesizer that obtains synthesized speech by filtering an adaptive excitation vector and a stochastic excitation vector stored in an adaptive codebook and in a stochastic codebook using LPC coefficients obtained from input speech. The gas diffusion portions and the spacer portions are integrally formed by bending a wire mesh member to have a rectangular corrugated plate shaped cross-section. A conversion dynode includes an ion collision region on a conversion dynode surface.