SPS-By-The-Numbers / sps-by-the-numbers

Repository for the website sps-by-the-numbers
MIT License
1 stars 0 forks source link

Try not to blind transcript readers with crazy background colors #12

Open msaxer opened 1 year ago

msaxer commented 1 year ago

The 10/17/23 transcript is a perfect example. The first 3 speakers (director Rankin, superintendent Jones, and Ms. Redmond) all have relatively bright backgrounds which makes reading tough.

I'm wondering if using colors from https://venngage.com/tools/accessible-color-palette-generator would be useful. Click "Load more palettes" and take a look at Pastel contrasting palette 1 & 2. Hit "Random palette" at the top until you get a set of colors you like.

Not sure if there's any clever way to prevent a separate issue in the transcript I linked to above where multiple speakers are assigned slightly different shades of green, which makes it a bit tricky to understand who is speaking.

msaxer commented 1 year ago

Looking at the Github code a bit I see you're using distinct-colors to autogenerate the color palette. I'm not sure how to force it to keep colors in its generated palette from being similar and non-blinding. Perhaps the "quality" parameter? Gotta confess I got lost trying to understand the linked "k-means convergence" on the distinct-colors Github page.

Perhaps creating a hard-coded palette of 50 colors which is used for all transcriptions would be best? That way you could choose readable colors and the ordering of the palette could keep similar colors (mostly) separated in the transcript since you can assume the first 10-15 speakers will show up most frequently so making their colors as different as possible might help.

One other color issue I can think of is color blind readers. Maybe IWantHue could be useful in that regard?

awong-dev commented 1 year ago

Thanks for the suggestions! This is definitely high on the list of bugs to fix as it is very much an accessibility issue as well as an aesthetic sanity issue.

We were also thinking a hard coded palette, but also possibly labeling folks by something other than just colors.

There's actually a whole accessibility issue with the DOM right now not having aria types or anything. If you felt like submitting pull requests (or jumping on the slack), patches totally welcome! Otherwise will try to get to it over the next couple of weeks...