Asterisk can perform text to speech using the Festival speech synthesis system.
The command Festival() provides a convenient interface to Festival. There is a
downside to this though. On a slow or busy system this can result in speech
that is garbled. For example the Pentium 200 system I run Asterisk on at home
drops most of the text that is sent to it and never completes a single word.
Using the text2wave utility that comes with festival you can solve this
problem. test2wave, as its name suggests, converts text into a wave file. It
normally reads text from STDIN and writes wave audio to STDOUT. A few command
line options make this process simpler. For example this command will convert a
text file into a wave file Asterisk can handle "text2wave -f 8000 -scale 5 -o
speech.wav text.txt". The "-scale 5" option increases the volume of the
resulting audio and might need to be adjusted on your system. The resulting
wave files which can then be placed in the Asterisk sound directory for
immediate playback at any time. For added space savings the wave file could be
converted to a gsm file using the SoX utility and the command "sox file.wav
file.gsm".

Another option is to create a script that performs this task at run time, like
festival-script.pl. This requires that Festival and the Perl AGI interface be
installed. I have modified that script to be more configurable and to cache the
wave files it creates. My version is named festival.pl and can also be called
via a symbolic link as cache-festival.pl. Both work by creating files based on
the MD5 sum of the text to be spoken. If called as festival.pl it creates the
files temp_[MD5].txt and temp_[MD5].wav in the directory set at the top of the
script. The files will be removed after the wave file is played. Should
Asterisk or the script crash before the files are removed they are easy to
identify and clean up later. If called as cache-festival.pl it creates the
files cache_[MD5].txt and cache_[MD5].wav in the directory set at the top of
the script. These files are never removed. The text file is left so that the
contents of the wave files can be easily identified later. In either case if
the wave file already exists the creation process is skipped and the existing
wave file is played. There are three potential problems with this script. The
first is if two different text strings have the same MD5 sum. While this is
exceptionally rare it is a possibility one should be aware of. The second is if
the script is called twice with the same text close enough together that they
interfere with each other. This could be solved by adding some sort of locking
mechanism. On a seldom used Asterisk system, like mine, this is not a problem.
The third is if the caller enters any digits while the wave file is being
played. Playback will be stopped and the digits will be lost. If you want to
Asterisk to capture the digits you must generate the wave file separately and
then play it back within the dialplan.

To install the script download it and copy it to the Asterisk AGI directory,
normally "/var/lib/asterisk/agi-bin", as festival.pl. Make any necessary
customizations to the variables at the top of the file. Create a symbolic link
for the caching version by executing "ln -s festival.pl cache-festival.pl" in
the AGI directory. Make sure the execute flag is set by executing "chmod +x
festival.pl" in the AGI directory. Finally create the directory to store the
temporary and cached wave files, this defaults to
"/var/lib/asterisk/sounds/festival/". Make sure Asterisk can write to and read
from this directory. Some systems run Asterisk as a non-root user for security
purposes. Utilizing this script is as simple as adding the following to an
extension "AGI(cache-festival.pl,'Welcome to Asterisk')". The script can be
found on my SDF Asterisk page.