@climbthemountain basically, I did what you do. On the image, I would read the numbers and what I think are important words. Since I believe nga, voice recognition Lang to, meaning the machine would just look for keywords, most of the keywords should be on the image. So I read what I think is important. But I do say, pie chart if it's a pie chart, line graph of its a line graph. Increasing if the trend is increasing. Eto hula ko Lang ha, if you say "the sun is yellow" or say "yellow sun is the", I think the machine will read it the same way. Most of the images, di ko natapos description ko. All I can suggest is, get as much as clear words as you can in. Confident Ako to give you an advice dito, kasi this part is where I scored the highest