There is a lot of success in translating speech into a foreign language speech in near real time. The next step would be to convert heavily accented speech (example: English) into normal pronunciation speech (example: English). This is called "auto-correct for accents". This would make webcast conferences (example: Google IO) more comprehensible.