Hear the body text, formulas and all

Add a PDF, HWP or HWPX document and DocVoice marks on the page what it will read, skips what gets in the way of listening, such as tables, page numbers and footnotes, and saves the reading as an MP3. Formulas are read structure first, so a listener can picture them. Everything happens on your PC.

Ask about the release

Coming to the Microsoft Store · price announced at release · Windows 10 and 11

The first page of the sample document Quadratic Formula Notes — The top of the first page of the sample document Quadratic Formula Notes. The sound is the MP3 this program made from it, unchanged (voice Female 5, speed 1×); while it plays, the highlighter colours the part being read. The tabs beside the page give each part's kind and reading time.

What it reads and what it skips

Read

Body text · paragraphs joined, as one sentence across a column break
Headings · with a pause, as headings
Formulas · display formulas recognised from the page image, then said structure first
Text boxes · like body text

Skipped

Tables · text broken into cells is hard to follow by ear
Page numbers, headers, footers · repeated on every page
Footnotes, captions · they interrupt the body text
References, contents · lists

Choose for each kind on the first screen whether it is read, and change single parts on the page once the document is open.

Formulas you can picture

A display formula is recognised from the page image by a formula recognition model; short formulas inside sentences are read by rules. Formulas are said a little slower than the text, with a pause at each symbol. This is what the program says for the formulas of the page above.

The quadratic equation and the quadratic formula on the sample page

Equation 1. a x squared plus b x plus c, equals zero. Equation 2. x equals a fraction. The numerator is minus b, plus or minus the square root of b squared minus 4 a c. The denominator is 2 a.

Your documents stay on your PC

Speech synthesis, character recognition, formula recognition and the language model all run on this PC. Documents are not sent over the internet, and the program needs no sign-up. A copy from the Microsoft Store only checks with the Microsoft Store component of Windows whether it was bought.

The program

The first screen: drop or paste a document, choose what to read, the voice and the speed.

The document screen: each page shows what is read; click a part to hear it.

What you need

System: Windows 10, Windows 11 (64-bit)
Disk space: about 4 GB
Memory: 8 GB recommended
Documents: PDF, HWP, HWPX (scanned PDFs without a text layer are read with character recognition)
Voices: 10 AI-synthesized voices (5 female, 5 male)
Speed: 0.8× to 5×, pitch unchanged
Screen: Korean or English · text size from 16 pt to 28 pt

Questions

Does it need an internet connection?

Not to read documents and make MP3 files. A copy from the Microsoft Store uses the Microsoft Store component of Windows only to check whether it was bought.

Does it read scanned PDFs?

Pages without a text layer are read with character recognition (OCR), which can make mistakes.

Does it always read formulas right?

Not always. Click the part on the page to hear it, and take it out of the reading if needed. Formulas whose characters are garbled are not read.

Can I use it with the keyboard and a screen reader?

Move between parts with ↑ and ↓, listen with Space, skip a part with Delete and save with Ctrl+S. Buttons carry names a screen reader announces, and text can be enlarged to 28 pt.

May I give the MP3 to others?

Please tell them it is an AI-synthesized voice; the MP3 file information says so too. The license of the voice model (OpenRAIL-M) forbids using it to impersonate or deceive others.

Who made it, and contact

I made it so that documents that are hard to read with the eyes can be listened to from start to end. It shows on the page what it will read, and says formulas in words. Tell me what to improve and it goes into the next version.

Geon Byeon · MarchToMay

Questions, bug reports, reports about readings made by the AI and requests for the source code of the open-source (LGPL) components included in the program: support@marchtomay.net

Seller MarchToMay (representative: Geon Byeon)