Thanks for all.
The B&K and GRAS mouth simulator are expensive.
I find a book wrote:
In certain cases, other source directivity patterns than omnidirectional can be of interest. For example, a directivity like that of a human talker may be relevant if the measurement is related to the intelligibility of speech or to measurements on the stage in an opera house. A suitable talker/singer directivity can be obtained by using a small loudspeaker unit (membrane diameter about three to four inches) mounted in a closed box of size comparable to that of the human head.
And in some papers, such as the PSB alpha Mite and PSB alpha II ( not sell now) loudspeakers are used for intelligibility test. From http://www.auditory.org/mhonarc/2005/msg00447.html, Franck used JBL: LSR25P for room acoustics.
I would like to use a small loudspeaker( it is cheaper than mouth simulator) to reproduce test signal. However, I do not know which types and brands of loudspeakers are suitful for my work. It may be best that the loudspeaker has its directivity data.