Reminder: OCR is the beginning and not the end of making your text accessible. It often outputs some garbage and always benefits from a look over and some cleaning up by a human. It is not magic and it doesn't mean there's no work left for you. Sorry.
@hollyamory People can DM the picture to OCRbot then use it as an image description directly on the picture once they've cleaned it up
As a woman that has studied machine learning since 2014 when time has permitted it, I can entirely verify this.
My IPA recognition algorithm is yet very naive and yet makes mistakes that it's training sets should have prevented despite the outcome of the training sets saying it.
That's after experimenting with different ways to numerically express and transform the sound waves into approximations to match what is the average of each syllable for highest reliability.
Pixels/shapes/shade combinations for training recognition is a considerable undertaking and is always skewed to the bias of the persons creating the training sets, which unfortunately lends to a need for oversight in Machine Learning to act against prejudices of the persons creating the training sets and overseeing said training.
Not only is it unreliable, it often times carries many prejudices to boot.
A community centered on the Twin Cities of Minneapolis and St. Paul, Minnesota, and their surrounding region.
An alternative to social networks that connected people in the region that have either died away or driven people off with unethical or anti-social policies.
MSP Social is a noncommercial community service, hosted and administered by @lawremipsum. Users should not feel obligated to contribute financially to the project. But contributions to defray server costs and/or for possible future expansion are welcome at Patreon, Liberapay or via PayPal to lawremipsum at gmail.
We financially support local community-oriented organizations. Currently, we support WedgeLive, Streets MN, Grease Rag and The MN Tool Library. Future support of community-oriented organizations will be determined by accountholders, donors and the admin, and is likely to be focused on groups that advance the values of the donors and encourage underrepresented voices in community and urban planning spheres.
Recurring contributors: @hfrazey, @firstname.lastname@example.org, @Jennybellium, @densetsu, @iangreenleaf, @britvulcan, @joeld, @yeahno, @paulference, @billmk, @brandon, @benjotron, @june @wafflesoup and anonymous.
If you're a current Twitter user, here is a tool that can help Twitter friends find each other on Mastodon.