Text to Speech That Reads Equations and Technical Terms
Most text to speech says 'r two' for R squared. Nook Listen rewrites notation, units and symbols into spoken words and skips citations.
Most text-to-speech tools are built for articles and audiobooks. Give them a research paper and they say "r two" for R², spell out backslashes in a LaTeX fraction, read every bracketed citation as a number, and turn an abbreviation like TGA into a made-up word. The audio is technically running and practically useless. Nook's Listen is built differently: before the voice sees a sentence, the text on screen is rewritten into the words a person would actually say. This page shows exactly how, with real examples.
The Idea: Change the Spoken Form, Not the Page
What you see on screen stays exactly as written. Only the text sent to the voice changes. That text is rewritten first, so "R² = 0.98 ± 0.02" is spoken as "R squared equals 0.98 plus or minus 0.02", while the page still shows the symbols and the highlight still follows the words as they are spoken.
Doing this well means knowing what a symbol means in context. The same mark can be a power, a charge, an orbital occupancy or a subscript, and it has to be spoken differently each time.
What It Sounds Like
| On the page | Spoken |
|---|---|
| R² = 0.98 ± 0.02 for n = 12 | "R squared equals 0.98 plus or minus 0.02 for n equals 12" |
| y = mx + c | "y equals m x plus c" |
| p < 0.05 | "p less than 0.05" |
| 12 wt% loss between 200 and 600 °C | "12 weight percent loss between 200 and 600 degrees Celsius" |
| Fig. 2 | "Figure 2" |
| TGA, SEM, FTIR | "T G A", "S E M", "F T I R" |
| ω, Δ, ≤, ≈, √ | "omega", "delta", "less than or equal to", "approximately", "square root of" |
| Φ with an "eff" subscript (LaTeX) | "phi effective" |
| N − 1 = k | "N minus 1 equals k" |
These cases come from Nook's own tests, not from marketing examples. Hear one for yourself. The text below has a unit, an abbreviation, a fit statistic, a figure reference and a citation marker:
It Reads the Same Mark Differently Depending on What It Means
This is the part generic text-to-speech cannot do. A superscript is not always an exponent:
- R² is a power, so it is "R squared".
- Fe²⁺ is an element with a charge, so it is spelled as an element and spoken with its charge.
- 1s² is an orbital with its occupancy, not "one s squared".
- H₂O is an element with a count, not a variable with a subscript.
- Mₙ is an identifier with an index, so it is "M sub n".
- A superscript number after a word is a footnote call-out, not a power, so it is skipped.
Chemistry, physics and statistics each have their own conventions, and the reader treats them as separate cases instead of one rule.
Citations and Footnotes Are Dropped
Bracketed references like [12] or [3, 14], parenthetical numbers after a word, superscript footnote marks and dagger symbols are removed from the spoken text. Without that, "prior work [12] showed" becomes "prior work twelve showed" and every few seconds a number lands in the middle of a sentence.
There is also a Skip notes option, which leaves out scholarly apparatus inside the body, such as figure captions, footnote bodies and reference lists, so you hear the argument and not the apparatus. It is deliberately cautious. If body text mentions a figure or says "note that", it is still read.
Tables, Equations and Code Get Their Own Handling
Tables are read with the headers once, then each row's values, so you hear a table as a table and not a stream of cells. The highlight follows the visible cells.
Displayed equations are handled as their own blocks, with the highlight following the lines.
Code listings keep their shape and work with the same highlight and reading tools as the rest of the page.
What It Does Not Do
Honest limits matter for a tool you will rely on:
- It does not describe a figure or a chart. Audio cannot replace looking at a graph. Pause, look, resume.
- It does not guess what an abbreviation stands for. If it does not recognize an initialism, it spells the letters out (T G A) instead of inventing an expansion. That is deliberate: an incorrect expansion is worse than a spelled one.
- It speaks the page, it does not explain it. To understand a passage, use Explain, which gives you a plain-language explanation of selected text and the definition of a single word.
- It is not claiming every field. The handling is strongest for mathematics, statistics, physics, chemistry and lab-science notation. If your field uses its own conventions, try a paragraph first.
Numbers in Other Languages
Listen covers 49 languages across 57 locales. For non-English voices, digits are spelled in that voice's language, so a Hindi or Spanish paragraph does not switch to English every time there is a number. English voices read numbers as they are, because they already read years, money and ranges well.
Why It Matters for Learning
For a dense paper, hearing the text spoken correctly gives you a second channel while you follow the highlighted words. Reading and listening together helps many people stay with the line and absorb dense sentences, but only if the audio is not fighting the content. A voice that stumbles on every symbol breaks your concentration faster than silence does.
Try Listen on a paper or import one now. The trial is 3 days with everything included.
How to Use It on a Paper
1. Import the PDF (or EPUB). Scans go through OCR on import.
2. Press Listen and start at 1x. Words highlight as they are spoken, and clicking a word jumps the voice there.
3. Pause on figures and displayed equations, look, then resume.
4. Use chunking for dense sections, so one paragraph is on screen while it plays.
5. Mark up as you go. Select a passage to highlight it or attach a note, and use Explain when something does not make sense.
For the full workflow, see How to Read STEM and Math-Heavy Papers on Screen and Best Text-to-Speech for PDFs and Research Papers.
Frequently Asked Questions
Can text to speech read equations?
Inline notation, yes. Nook's Listen rewrites symbols, operators, Greek letters, units and chemistry notation into spoken words before the voice reads them, so R² is "R squared" and ± is "plus or minus". A displayed derivation or a figure still needs your eyes.
Why do most text-to-speech tools fail on research papers?
They read symbols literally, say "r two" for R², spell out LaTeX, read citation numbers as part of the sentence and invent words from abbreviations. They are built for articles, not for scientific notation.
Does Nook explain jargon?
Listen speaks technical terms and notation correctly. For explanations, Explain gives you a plain-language explanation of any passage you select, and selecting a single word shows its definition.
Does it skip citations?
Yes. Bracketed references, parenthetical numbers, superscript footnote markers and dagger marks are removed from the spoken text. The Skip notes option also leaves out captions, footnote bodies and reference lists.
Does it work for chemistry and physics?
Yes. It reads chemistry and physics notation by context: powers, charges, orbitals, stoichiometry and units are each spoken the right way.
Which languages does it cover?
49 languages across 57 locales, with regional variants as separate voices.
Related reading:
- Best Text-to-Speech for PDFs and Research Papers: the tools compared
- How to Read STEM and Math-Heavy Papers on Screen: the full workflow
- What Is Nook Listen?: follow-along basics
- Nook Listen vs Speechify: same paper, two voices
- Best PDF Reader for Studying: the apps compared
- How to Read a Research Paper: pair audio with a method