Mary TTS is an Open Source, Free, Text to Speech Java based package that runs on multiple platforms and includes a facility to generate your own voice packages.
- You can potentially simply read and record a few hours worth of your favored type of material but then to follow the proces, here, you would have to split the recordings up into .wav files of a sentence or just a few words it does not work too well on long blocks and provide a matching transcription for each file and there is a reasonable chance that you would be lacking at least some phonomes.
- You can use a set of predefined training scripts, adapted from those provided by HTS as detailed here - this has the advantages of the transcriptions already being available, the fragments being the correct lengths & the phonome coverage being good but are not necessarily fun to read.
Looking at the process involved the majority of the tools required for voice creation are Linux or OS-X based and there is not much mention of Windows but once the voice is created you can transfer it to a Windows installation.
I have to admit that I do not know off of the top of my head the process for using it with an ebook.