Crowd Sourcing Transcription
Here are some ideas for how to crowd-source the editing of transcriptions. SpeechRecognition will leave us with a set of segmentations and imperfect transcriptions, which will be suitable for indexing but may or may not actually be comprehensible.
Nonetheless, imperfect transcriptions are a good basis for manual transcription, since it's much easier to fix an existing transcription than to create a new one from scratch.
Paul's idea is to show the transcriptions along with the videos with an interface such that people watching them can easily and quickly fix errors if they are so inclined.
User interface concepts
Have an auto-scrolling text panel next to the video - the "bouncing ball" approach. Clicking on a transcription segment either makes it editable, or brings up a bubble to allow you to "suggest" an alternative - much like Google Translate.