I banked my voice with ModelTalker. It was free.
It is sensitive to the microphone you use, as a certain quality level is necessary before a voice file can be generated.
I used a Sennheiser PC8 USB headset on Windows 7 Professional and it worked out very well.
The process is well-described on their web site.
You download some stuff, record a sample, and submit it for approval. The approval process is making sure you are able to record meaningful-enough voice patterns to eventually generate a workable voice file.
I failed at this initial step, purchased the sennheiser PC8 mentioned above, tried it again, and passed.
After I passed, I progressed to the step of recording a large number of phrases they provided. I did it first thing in the morning when my voice was fresh.
I followed recommendations for professional speakers, which include consuming no fat or dairy before recording the phrases. A warm beverage such as tea, black coffee, or just hot water can help prepare the voice for better speaking. These are tricks I used throughout my career, where I was called on to do quite a bit of public speaking.
Some mornings, I could only get through 15 phrases before my voice started to degrade. Other mornings, I could do as many as 100 phrases (though, most of them time it was 50 or less). I wanted the highest-quality sound file possible, so took care to re-record phrases where my voice had degraded. It took several weeks to do this task.
Once that was done, I submitted my recorded phrases for voice file generation. A bit after that, I received the file that is used to generate my voice. I backed that file up on 6 different disks and my cell phone. One disk is always stored off site.
I have installed the TalkTablet app on my android cell phone (Samsung Galaxy 4s). I can type in a phrase and then play it with something that sounds like my voice.
Some phrases sound just like me. Others not so much.
I don't need speech generation right now, but as my voice continues to degrade I know I will. Overall, I am very satisfied with ModelTalker and glad that I banked my voice using it.
Steve