Reported by Agran on 2012-08-06 11:30

1. Identify picture as diagramm, plot. (no clue, trial and error?, tagged?)
2. Extract scaling values (OCR)
3. (Dis)play plot, diagramm as time dependent audio.
4. Read out scaling with reference audio, repeat diagram
5. Repeat on demand

After recognizing the type of diagram, you should handle diagrams either as continuous diagrams like stock exchange curves or singular point or bar diagrams.

Continuous diagrams should have a short forte when crossing a y-scaling value and a metronome could give x scaling.

Bar diagrams are a bit easier to understand. After reading category you make the audio elevate from 0 to max y (including a forte on the scaling, which should make it more a bit easier), than next category.
You could probably set single points in categories instead of a bar-like elevation of the audio, but this seems more comprehensible.

You can even have parallel categories/curves "displayed" with different instruments/sound characters

I hope someone finds this useful, I hopefully won't need it. If I'd knew how to program it, I'd done it myself. But alas, I hope this helps and of course all the best luck for the recognition part which is quite complicated. I only realized this when I wrote down the step by step approach. I won't be here often, but feel free to contact me on the subject chrush at web dot de