Transcribing photographed typewritten archives: which OCR to choose?

Which OCR should I use for photographed typewritten archives?

For typewritten archives photographed in a reading room, the best-suited tool is one that reads a handheld photo, not only a flat scan. wondarchive transcribes this typewritten text with excellent accuracy and respects the page layout, including tables in columns. It then makes the collection searchable, translatable and exportable, with images hosted in France. A desktop OCR remains relevant for clean scans made on site.

wondarchive is an online service that transcribes, translates and indexes archive photos taken in the reading room (manuscripts, typescripts, printed documents), for historians, doctoral students, archivists and genealogists. The details of what is read, page type by page type, are on what wondarchive reads.

Why is a photographed typescript harder to read than a scan?

A typescript photographed in a reading room is harder to read than a scan because the photo adds its own flaws to those of the document: a slightly tilted page, uneven lighting, a two-page spread, the shadow of the binding. On top of these come the typescript's own problems: faded carbon copy, translucent onionskin paper, uneven typing, handwritten corrections. An OCR designed for clean, straight scans copes poorly with this combination.

• Shooting conditions: wondarchive turns a page rotated by a quarter or half turn the right way up before reading; a page tilted by a few degrees is read as it is.
• Two-page spreads: a bound report photographed open, in a single shot, is read as two pages, left then right.
• Sharpness: on a faded carbon copy, quality depends first on the photo; the photography guide comes down to six rules.

Which tool should I choose to transcribe typewritten archives?

Choosing a tool to transcribe typewritten archives depends on four criteria: the form of the images (reading-room photos or flat scans), the volume, the technical skills available and where the images may be sent. The table below compares five solutions against these criteria. For each, it gives its real strength and the case where it is the better choice.

The detailed comparison with desktop OCR is on ABBYY FineReader or wondarchive. The one with cloud APIs is on OCR hosted in France or US cloud.

How do I transcribe a typewritten collection step by step?

Transcribing a typewritten collection with wondarchive takes five steps, with no file preparation. You upload the folder of photos as the camera wrote it; the tool straightens, transcribes and describes each page; you then read, check and organise the collection. Transcription continues on the server, even with the tab closed, within minutes or hours.

1. Upload the mission folder, subfolders included; files already online are not sent again and the original JPEG is kept intact.
2. Straighten: the browser decides the orientation of each page; if in doubt, it shows the four orientations and you click the right one.
3. Transcribe: each page is read at full resolution; the tool also notes the type of document, its language, its date, the dates cited and the proper names.
4. Read and check: photo on the left, text on the right, original or translated in one click.
5. Organise: place, shelfmarks, folios; the shelfmark printed at the foot of the page is read and suggested for the following pages.

Are typewritten tables in columns preserved?

Yes: on a typewritten page, wondarchive respects the layout, including tables in columns, instead of flattening them into an unordered string of figures. This is useful for accounts, statements and statistical reports in twentieth-century business and administrative archives. The text of these pages is then searchable across the whole collection and exportable to PDF.

How do I check the transcription of a typescript?

A typescript transcription is checked by comparing the text with the photo, which wondarchive displays side by side: a proper name, a figure or a date can be verified at a glance. On typewritten text, accuracy is excellent; checking focuses mainly on proper names, amounts and passages where the carbon copy is faint. A correction is made in the text without losing the original version.

Can I search, translate and export transcribed typescripts?

Yes: once transcribed by wondarchive, typescripts are full-text searchable across the whole collection, translated page by page into your language and exportable. The translation takes into account the end of the previous page and the start of the next, and follows a glossary of the collection's terms. Search finds a name with its variants, or a date even if imprecise.

A page, a shelfmark or a search can be exported to PDF: see exporting pages.

Follow-up questions

What about very faint carbon copies or onionskin paper?

Reading a faint carbon copy depends first on the photo: a sharp image, page flat, as the photography guide explains. The safest approach is to try a few difficult pages from your collection within the 100 free pages.

Are handwritten marginal notes on a typescript read?

Yes: handwriting, marginal notes included, is transcribed with reinforced proofreading. Quality is high, but checking against the photo is still advised, especially for names and initials.

Which languages can typescripts be transcribed in?

They can be transcribed in European languages, including German, French, English, Italian, Spanish, Dutch, Latin and Russian, in Latin, Cyrillic and Greek alphabets. The text is transcribed in its own language and translated into yours.

Do my archive images stay confidential?

Images are stored in France with OVHcloud, read in the EU, are not used for any training and can be exported in full. Who reads them and what is not done with them is detailed on your data.

Is a desktop OCR enough if I have clean scans?

For sharp, flat scans, a desktop OCR or an open-source engine may be enough to produce text. wondarchive adds reading of reading-room photos, translation, search across the whole collection and organisation by shelfmark.

Try it on your own typescripts

The simplest way is to upload a few dozen photos from your typewritten collection and compare the text with the image. Opening a free account gives 100 pages transcribed and translated, with no credit card. To see real archives being read first, rights-free, open the demo.