i think TTS is promising and the API is neat, but this project, due to it's educational nature, should be using good role models. especially on single words where context doesn't provide any clues... TTS looks great for other accessibility issues like reading aria descriptions and announcing subtitle descriptions on video. besides, they already have the recordings and it's a known set of so many, not a bottomless vocab like you would find in subtitles.
BROWSER STATS [% share]
(2013/10/31) IE7:0.5, IE8:8.6, IE9:5.3, IE10:12.3, FF:17.7, CH:41.8, SF:8.1, MOBILE:20.4