Thanks for all the advice. I don't have Windows, so CoolEdit is out of the question. I've had a play with Festival, generating WAV output and converting with sox:

all: $(subst .txt,.pcm,$(wildcard *.txt))

%.wav: %.txt
[tab] /usr/bin/text2wave $< >$@
[tab] /usr/bin/normalize $@ 2>&-


%.pcm: %.wav
[tab] /usr/bin/sox $< -t raw -r 44100 -w -c 2 $@ resample -qs

clean:
[tab] rm -f *.pcm *.wav

The speech quality is quite a long way short of the distributed TTS Clock, though - adequate for reading emails, say, but not for something I hear every day. I think I ought to investigate the alternative of persuading a friend to record the samples. I don't have studio-quality (or in fact any) recording gear, though, so it may take a while.