PDF & photo scanning: what's new.

You are reading it. Everything that reaches you, from new features to smaller fixes, is listed here in your language, with a link to the help centre whenever a page exists. Find it under Product Updates in the site menu: flat.io/changelog, flat.io/edu/changelog and opuscan.com/omr/changelog.

A page engraved with smaller or larger staves no longer comes out at one fixed default size.

Stanzas printed apart from the music, copyright footers, tonic sol-fa rows and fixed-do note names are no longer sung as lyrics. Horns labelled "Hn." or "Horn 1" now transpose in F, choirs labelled S, A, T, B and claves are recognised, and credit labels in French, German and Korean fill the right fields.

Pages with many staves, from orchestral full scores to marching band charts, are no longer split into extra systems or merged into one. Time signature changes printed on a single staff, multi-bar rests and first endings retain their bars, and very tall stitched captures and skewed photos are read more reliably.

Some PDFs store their Japanese text in an encoding that used to come out as unreadable characters. Lyrics and titles from these files are now read correctly, and an import set to several languages now uses all of them.

Tied notes now reach their second note in other notation apps, and a 22-bar rest is exported as 22 bars instead of one, so bar numbers match the original.

On clean, typeset pages the page analysis sometimes hesitated between two nearly identical outlines of the same system, and the split confidence let the system slip through unrecognised. A very faint staff sitting just under the detection threshold could go the same way. On our scanned-page benchmark, the share of systems found rises from 96.1% to 99.1%, with single-line parts gaining the most (90.6% to 97.8%).

A chart containing only plain chord symbols (C, F, G) is now read as chords too, instead of having them sung as lyrics.

A page whose parts are labelled Soprano, Alto, Tenor and Bass is now recognised as a choir, so the import no longer stops to ask you to confirm the three lower voices.

Photographing a monitor or tablet adds a fine pixel grid over the page that used to hide the staves. Imports now see through it.

Pages with bleed-through, duplicate detections, a changing layout or a shorter appendix keep all their systems. A misread can no longer add hundreds of empty bars, and cropped excerpts are no longer flipped upside down.

Clefs and figured bass appear only on the staves that carry them, pickup bars are numbered as in print, multi-bar rests and trill lines export as valid MusicXML, and cross-staff piano voices stay on the correct staff.

Marks such as "1 Verse" or "Level 2" are no longer dropped, and stray letters no longer become a rehearsal mark.

Choir abbreviations, "Violin II", Japanese labels and boxed section names are handled, stray glyphs no longer become part names, and the instruments you confirm before import are always applied.

Chords over rest-only bars are kept, chords split across two text boxes or written with the alteration after the degree (G7⁵⁻) are read, and a row of bare chords is no longer sung as lyrics.

Copyright lines, stamps, section labels, mood words, narration cues and page headers stay out of the title and composer fields. Japanese credits are read with their role label, and a title is never duplicated.

Syllables stay on their notes and in their bar, the first and last syllables of a line are kept, verse numbers and drum sticking are not sung, and lyrics printed between two choir staves go to the right one.

A mark clipped at the top of the page, written inside a sentence, or glued to a rehearsal letter is now read. Relative directions such as rit. or accel. no longer invent a tempo.

Metronome equations and tempo words such as "poco allegro" stay where they belong.

Credits in Polish and Spanish, wrapped over two lines, split across text boxes, or set in the music font are now recognised.

• A progress bar while a scan is exported.
• A failed import offers to send a report instead of opening the help centre; the help link points to opuscan.com/support.
• App version and support details in Settings.
• The credit paywall only appears when credits are the reason.
• Guests keep their identity across sessions.

If you close the review screen or lose the tab, the import no longer waits forever: after a day it continues with the title and instruments it detected, and the score lands in your library.

The import flow on Android has been rebuilt on the same engine as the web app: pick the pages you want, check the detected title and instruments, then let the scan finish while you keep using the app. Opuscan for Android gets the same wizard.

In some piano scores, the second voice inherited the staff of the previous one, so right-hand notes were written on the bass staff.

Busy systems that continue from the previous line no longer lose notes, a part that mixes triplet and plain systems no longer turns into rhythm slashes halfway through, and slurs and articulations placed on a double stop stay on the chord.

Text recognition now covers Vietnamese, Thai, Arabic, Persian, Urdu, Hindi, Marathi, Nepali, Bengali, Tamil, Telugu and Kannada, plus 24 more Latin-script languages such as Czech, Hungarian, Croatian and Lithuanian. Words with accents and diacritics are also kept instead of being discarded as unreadable, and Hebrew text embedded in a PDF is no longer thrown away.

A bare "Horn" now resolves to the French horn instead of the cor anglais, "Alto" and "Tenor" on a band part to saxophones, brass band basses to the tuba, and a percussion clef misread on a pitched part is corrected. A vocal line that enters after a piano introduction is recognised as a voice instead of a second piano. On Chinese scores, the 詞 and 曲 labels route the names to lyricist and composer, and bilingual lyrics keep every verse. Changing the instrument of a drum part in the review step keeps the drum kit.

Photograph an open book and the import reads the left and right pages one after the other, instead of as one wide page.

If you run out of credits partway through a scan, you can buy a pack and the import carries on from where it stopped. A confirmation appears after each purchase, and Help and support opens the new Opuscan support site.

Chord symbols written with solfège names are recognised, extensions stacked over the chord ("sus4" under "A7(6)") are folded back onto it, and sharps and flats drawn in the notation font of a digital PDF are read instead of dropped. Chord symbols detected by text recognition are imported as chord notation rather than discarded.

Page and catalogue numbers no longer shift the bar numbering, "Words and Music by" and bare "by" lines no longer end up in the composer or subtitle, flush-left movement headings go to the movement title, boxed rehearsal numbers are kept, and expression text above or below the staff is imported as a direction rather than as a credit.

First verses and second-verse syllables near a neighbouring staff are recovered, the sung line no longer scatters across both staves of a grand staff, and copyright footers, end-of-piece credits, tempo words, stage cues and stray music-font glyphs are kept out of the lyrics.

Part names printed in the margin or in the header now match the right instrument more often for organ, choir, percussion and band scores, including abbreviations, plurals and ordinals ("1st & 2nd Horns in F"), and names in Japanese and other languages are looked up too. Dynamics, metronome marks and lyric syllables next to the staff are no longer mistaken for a part name. Transposing instruments are exported with their transposition so they play back at concert pitch.

A pickup bar is carried to every part instead of being created from a lone short bar, printed multi-bar rest counts are read as such, bars that overflow by one duration are corrected from the model's next best guess, and notes with a missing duration inherit the current one instead of disappearing. Unpaired ties and tremolos or noteheads the editor could not load no longer make the whole import fail.