I hope someone submits a voice recognition data!

It's fantastically hard to find audio->phoneme datasets and only slightly less difficult to find audio->word and word->possible_phonemes...