"100kg" and "2024/12/31", read correctly — Yomite's number, unit, and symbol normalization
Most TTS apps read "100kg" as "hyaku kē jī" and spell out dates digit by digit. Here's how Yomite's dictionary-based text preprocessor normalizes dates, units, currency, math, chemical formulas, Roman numerals, and Greek letters before synthesis — with real before/after examples.
“100kg” read out as “hyaku kē jī.” A date like “2024/12/31” spelled out digit by digit — “two-zero-two-four slash one-two slash three-one.” “H₂O” trailing off after just “H.”
If you’ve used a system TTS voice or a generic cloud TTS engine, you’ve probably run into this. Text with numbers, units, and symbols mixed in is a minefield for most speech synthesis.
Yomite solves this with a dictionary-based preprocessing layer. Before any text reaches the synthesis engine, it normalizes dates, times, currency, SI units, math operators, chemical formulas, Roman numerals, fractions, and Greek letters into the way a person would actually say them out loud.
What actually changes
Yomite’s site publishes a comparison table with real conversion examples. Here’s a slice of it:
| Input text | Other TTS (system / generic cloud) | Yomite |
|---|---|---|
2024/12/31 | “two-zero-two-four…” | “December 31, 2024” (read naturally) |
12:30 | “one-two colon three-zero” | “12:30” (read as a time) |
100kg / 3.5m² / 30°C | “hyaku kē jī” / skipped | “100 kilograms / 3.5 square meters / 30 degrees” |
¥1,000 / $50 | “one-zero-zero-zero” (symbol skipped) | “1,000 yen / 50 dollars” |
3 + 4 = 7 | “three four seven” (operators ignored) | “3 plus 4 equals 7” |
H₂O / CO₂ | “H” / “C-O” only | “H-two-O” / “C-O-two” |
½ / ¾ | often skipped entirely | “one half” / “three quarters” |
α, β, γ | misread, not recognized as Greek letters | “alpha, beta, gamma” |
The categories covered fall into eight groups:
- Dates & times (
YYYY/MM/DD,HH:MM:SS) - SI units (kg, km, Hz, °C, m², kWh…)
- Currency (¥, $, €, £)
- Math operators & symbols (+, −, ×, ÷, ±, =, ≠, %, ‰)
- Chemical formulas (H₂O, CO₂, NaCl)
- Roman numerals (Ⅰ–Ⅻ)
- Greek letters (α, β, γ … ω)
- A standard dictionary of 5,000+ entries covering commonly mispronounced words and web-novel/light-novel proper nouns
These conversions are continuously verified by automated tests and applied automatically every time text is loaded — there’s no setup required on your end.
Why this needs no configuration
Yomite has a separate feature: a per-book pronunciation dictionary where you register custom readings for things like invented fantasy names that don’t exist in any dictionary.
Number, unit, and symbol normalization is a different kind of problem. How “100kg” should be read isn’t something that varies by book — it has one correct reading in natural speech. That’s why Yomite builds this in as a standard, always-on layer rather than something you configure. It applies the moment text is imported, with nothing for you to set up.
Where this actually matters
Normalization pays off most outside of fiction, wherever numbers and symbols carry real information:
- Papers and technical documents — when you listen to PDFs and papers, chemical formulas, units, and equations show up constantly. If “H₂O” or “100GB” gets mangled every time, comprehension suffers.
- Business documents and reports — dates, timestamps, amounts, and percentages need to come through correctly the first time, or you end up rewinding to catch the number you missed.
- News and web articles — statistics, prices, and temperatures are often the actual point of the sentence.
- Recipes and how-to guides — measurements (½ cup, 100g, 30°C) are everywhere in this genre, and a skipped unit changes the meaning of the instruction.
The takeaway
Getting distracted by “hyaku kē jī” instead of the content itself is a small thing that adds up to a real dent in immersion. Yomite normalizes dates, times, units, currency, math, chemical formulas, Roman numerals, Greek letters, and a 5,000+ entry standard dictionary automatically, through a dictionary-based preprocessing layer — no setup required. Import your text and it reads naturally from the first sentence.
If you regularly listen to text full of numbers and symbols, the difference is audible immediately.
Get 読み手 Yomite on the App Store →
Related: Fixing TTS mispronunciations with the pronunciation dictionary · Listening to papers and PDFs
Questions or feedback: [email protected]
— Yomite Team