I think TTS is not a good choice. TTS won't let you learn nuances and accents.
In case of IVONA TTS it strongly depends on the word itself. The wast majority is pronounced flawlessly. I was quite skeptical myself about possible use of TTS generation until I found some real things. Any TTS generator is not good to voice, say, a book, but a good generator is more than enough when you want to voice a word combination or a simple sentence. In this case even the intonation is rather appropriate. If you doubt it, then just check it out - there are fee fully functional demos.
I don't mean that a voice generator is better anyhow than an authentic audio record, but there are a lot of cases when words are not voiced at all. No word combinations are voiced as far as I know, if, certainly, we are not speaking about the LDOCE from the presently dead traduko site.