Instructions

The steps

  1. Press "Choose a file" and pick a photo, image or PDF. On a phone you can take the photo there and then.
  2. If the file has several pages (PDF or TIFF), enter the page you want, or choose "All pages" for a single Braille file, up to 40 pages at a time (about 15 seconds a page). Each page of the book starts on a new Braille page.
  3. Press "Convert to Braille". Recognition usually takes 10–30 seconds; progress is announced every few seconds.
  4. Under "Result", download the file you need or copy the Braille.
  5. Check the result. See accuracy below.

How to take a good photo

The quality of the photo matters more than anything else. Even a slight tilt or blur can spoil recognition a great deal.

Tips for blind users
  • Lay the open book on a table near a window or under a lamp.
  • Hold the phone with the camera facing down, about 30–40 cm above the centre of the page.
  • VoiceOver and TalkBack often announce when text or a document is in view, which helps to centre the page.
  • If the app finds very few signs, it retries by itself with straightening and clean-up. If there are still few, it tells you why (for example blur or poor light); then take a new photo.

Which files it accepts

The results and what to do with them

FileWhat it is forHow to use it
.brf (Braille Ready Format)EmbossingOpen it in your Braille embosser's software and emboss. It is paginated at 40 cells × 25 lines, or the width you chose.
eBraille (.txt)Braille display, electronic readingOne character per cell (Braille ASCII). In NVDA choose the output table "English (U.S.) 6 dot computer braille" so that each character shows as one cell.
Unicode Braille (.txt)Viewing on screen, email, documentsThe dots are visible on screen; useful for a sighted teacher.
.byzocrWhat was recognisedOpens in the Neanes program for correction; you can then upload it here again.

In the "Recognition check", each Braille segment is listed with a description of its signs, for example "petaste + oligon, klasma". When you select a segment, the signs it came from are highlighted on the page; signs with low confidence have a dashed orange box. Warnings mark places the Dositheos method does not cover explicitly: there the converter does not guess, and a human check is needed.

Correcting errors and adding lyrics

Automatic recognition makes mistakes. Anyone who can see the score, a teacher or a chanter for example, can correct them in the page, without knowing Braille.

  1. Under "Check and correct", press "Correct" next to a segment, or click a sign on the page.
  2. For each sign, change what is needed: neume, time sign, gorgon, sharp or flat, psephiston or antikenoma, vareia, phthora. For martyriae you change the note, register and genus.
  3. "Delete this sign" if the sign is not in the score, "Add a neume after this one" if one is missing.
  4. Lyrics: in the "Syllable" field of the neume on which each syllable starts, type the syllable, e.g. "Κυ". The Braille then writes the syllable with its music, as the method prescribes.
  5. Press "Apply corrections". The Braille is rewritten at once by the Dositheos rules, and corrected signs get a green box.

There is Undo and Back to the original reading. Results are deleted after 6 hours, so press "Save corrections" to keep your work in a file (.byzdots.json). Upload it again to carry on where you left off. The page image is not stored in that file.

The editor works with the keyboard alone and with screen readers, so a blind user who knows what the score says can use it too.

Accuracy and limitations

Conversion happens in two stages, and errors can come from either:

  1. Recognition (OCR): the signs are found in the image by our own recognition engine, which runs inside the application with no other service. It recognises about 99% of the pieces of ink correctly on pages it did not see in training.
  2. Conversion to Braille: done with the rules of the Dositheos method from the thesis. On the thesis's four worked examples, a cell-by-cell comparison with the thesis's solution, from the photo to the Braille, gives a similarity of 89–97% (Kekragarion in the grave mode 97%, Kekragario in the first mode 95%, Pasapnoario in the plagal second 94%, Agni Parthene 89% from a very small image). Most differences are martyriae and rare signs.

What this means in practice: on a clean page with 100 signs, expect up to about 10 errors, and more on blurred or tilted photos. That is why the Braille must be checked by someone who knows the method before it is used for teaching or at the chanter's stand.

Known limitations:

How to find errors: the martyriae are checkpoints. If a martyria is not the one you expect, the error lies somewhere between it and the last correct martyria.

Frequently asked questions

Does it cost anything?

No. The service is free and needs no account.

Do you keep my photos?

No. They are deleted automatically within 6 hours and used for nothing else. See the privacy policy.

Very few signs or none came out. What do I do?

Take a new photo: closer, with more light and the page straight. If you have a PDF, use it. The app already tries straightening and clean-up by itself, and tells you if the photo looks blurred, dark or faint.

Is the Chrysafidis method supported?

Not yet. The output follows the method of the monk Dositheos, approved by the Holy Synod in 1960.

Can I correct a sign in the app?

Yes. Press "Correct" next to the segment, or click the sign on the page. See correcting errors.

How do I add the lyrics?

In the editor, type each syllable in the "Syllable" field of the neume on which it starts. You can also ask for experimental automatic reading of the lyrics when uploading; then check them in the editor.

Does it work with my screen reader?

It was designed for NVDA, JAWS, VoiceOver and TalkBack. See the accessibility statement and let us know if anything does not work.

Acknowledgements

Recognition uses our own engine. The free fonts of the Neanes team (SIL OFL 1.1) were used to measure how signs combine and for synthetic examples; at the start, the open-source Neanes byzantine-chant-ocr gave the first proposals for the training labels, each of which was checked. We thank papline editions for permission to use their pages in training. A key work in the field is C. Dalitz, G. K. Michalakis and C. Pranzas, "Optical Recognition of Psaltic Byzantine Chant Notation" (IJDAR, 2008). The Braille rules come from the doctoral thesis of Fr. Theodoros Tsampatzidis. All third-party work is listed under software licences.

Questions or problems: info@byzantine-dots.gr