They should just make them available with the sound all distorted and super noisy like partially deaf people face, or just no sound, and autogenerated subtitles

They should just make them available with the sound all distorted and super noisy like partially deaf people face, or just no sound, and autogenerated subtitles. Then, have online students sign up to write sub-titles or verify subtitles. For every hour of subtitling you get 20 hours of hearing. Verification via a CAPTCHA like system that interrupts video playback to verify snippets of audio. Sell this corpus to deep learning people. With hundreds of thousands of people doing MOOCs it should be very doable.