How We Pronounce Your Name

English speakers know this intuitively: "cough," "through," "though," and "dough" all end in "-ough" but rhyme with nothing. Names are no different. "Sean" is pronounced "shawn." "Cholmondeley" is pronounced "chumley." Spelling alone is unreliable.

For Japanese calligraphy, we need actual sounds — not guesses from letters. A system that converts "Sean" letter by letter would produce something unrecognizable. We need to know that it sounds like "shawn" before we can render it faithfully in Japanese.

For an overview of how Japanese characters represent those sounds, see How to Write Names in Japanese.

Five Pronunciation Authorities

We do not guess how names are pronounced. Every pronunciation in our system traces to at least one authoritative dictionary.

CMUdict (Carnegie Mellon University Pronouncing Dictionary) — 134,000 English words and names with precise phonetic transcriptions. Created by speech scientists for speech recognition research. When we know how Katherine is pronounced in English, CMUdict is typically the authority.

MFA (Montreal Forced Aligner) — 3.5 million entries across 12 languages. The broadest multilingual resource. MFA knows how Guillaume sounds in French, how Friedrich sounds in German, how Bjork sounds in Swedish.

IPA-Dict — One million entries across 31 languages, with full stress marking. When we need to know that Eduardo stresses the third syllable, IPA-Dict provides that precision.

Census IPA — 147,000 name-pronunciation pairs from government census data in 14 countries. These are names that actually appear in government records, cross-referenced with pronunciation dictionaries.

BehindTheName — Pronunciations from native speakers for over 65,000 names. A BehindTheName pronunciation of an Irish name comes from someone who knows Irish naming traditions, not from a generic phonetic algorithm.

Together, these five sources contain over 4.5 million pronunciation entries spanning 31 languages. The system searches them in sequence — most authoritative first, broadest last — to find the best available pronunciation for each name.

Every Language Has Its Own Rules

The challenge multiplies across languages. In German, "W" is pronounced like English "V." In French, most final consonants are silent. In Finnish, double consonants are always long. In Polish, "sz" is a single sound, like English "sh."

Our system handles names from over 30 language backgrounds. Consider these names, all written in Latin letters, none with reliable English pronunciations:

• Paivi (Finnish) — the "a" is a specific vowel not found in English
• Sian (Welsh) — the circumflex marks a long vowel; the whole word is one syllable
• Grazyna (Polish) — the "z" with a dot is a specific consonant, not a "z" with decoration
• Olafur (Icelandic) — the accent marks are phonetically meaningful

An English-only system would mispronounce all of these. Multi-language dictionaries are a necessity for serving a global customer base.

The English bridge

Until full multi-language support is complete, names from unsupported languages receive English-approximation pronunciations. This is not ideal — a German name processed through English will not capture German vowel qualities precisely — but it is a pragmatic bridge. The original language is preserved, so when native-language processing becomes available, the pronunciation can be upgraded.

How Stress Shapes Your Name

Where you place emphasis changes everything. "ANdrea" and "anDREa" are different pronunciations of the same spelling, and they produce different Japanese renderings.

Stress matters for Japanese because emphasized syllables typically become long vowels — they get more time and space in the final art. An emphasized "AH" becomes a full, prominent sound; an unstressed "uh" becomes a quiet, short one. Where stress falls literally changes the shape of the calligraphy.

Different dictionaries record stress differently — CMUdict marks every vowel with a stress level, IPA-Dict uses standard phonetic markers, MFA provides sounds without stress marking at all. When the primary dictionary lacks stress information, the system consults other dictionaries that do provide it.

When our dictionaries record multiple stress patterns for the same name, both become separate pronunciation variants — each producing a different Japanese rendering that you can choose between.

Native-Language Pronunciations

A name's truest pronunciation is how the people who gave it say it. "Jorge" to an English speaker might be "jorj." To a Spanish speaker, it is "hor-hey." To a Portuguese speaker, it is "zhor-zhee."

For over 15,000 names in our system, we have pronunciation data from BehindTheName that reflects how the name sounds in its native language — not just an English approximation. The BehindTheName pronunciation captures the Spanish or Portuguese rendering: the one a customer named Jorge most likely grew up hearing.

Census data provides a different kind of authority. When the same name appears in government records from France, Germany, and Brazil, that is evidence of genuine international usage — and each country's pronunciation dictionary provides the locally correct rendering.

This enrichment is additive. The English pronunciation stays, and the native-language pronunciation appears alongside it as an additional variant. You choose which rendering you prefer.

When We Need Your Help

Sounds that Japanese cannot distinguish

Some sound distinctions that matter in English do not exist in Japanese. Japanese has one sound that sits between English L and R. "Larry" and "Rally" produce identical Japanese characters. But the pronunciation guide should show "L" for Larry and "R" for Rally.

Similarly, "S" and the voiceless "TH" merge in Japanese — "Seth" and a hypothetical "Sess" would look similar. So do "D" and the voiced "TH" (as in "the").

In most cases, your name's spelling resolves the ambiguity without asking. "Yilmaz" contains an L, not an R — the spelling makes it obvious. "Ruth" contains a TH, not an S. When the spelling unambiguously answers the question, the system silently uses the correct sound.

We only ask when the spelling genuinely does not resolve it — and the goal is to ask as rarely as possible. Most names resolve automatically. But when ambiguity is genuine, asking is better than guessing wrong.

Entering Your Own Pronunciation

Dictionaries cover the most common names comprehensively, but the long tail of rare names, family-specific spellings, and cultural variants means some names are not in any dictionary.

Phonetic text

You can type your pronunciation using English phonetic spelling — writing it the way it sounds rather than the way it is spelled. "Catherine" might be entered as "KATH-er-in" or "kath-RIN" depending on your accent.

The system parses this input, identifies consonant-vowel patterns, and determines stress. When something is ambiguous — "did you mean the 'oo' as in 'food' or as in 'book'?" — it asks rather than guesses.

Direct romaji

For those who already know the Japanese romanization they want, the system accepts direct romaji input. Typing "maikeru" directly tells the system exactly which Japanese syllables to use.

The input is validated against the full syllable inventory — 453 recognized Japanese syllables. If the input contains an invalid syllable, the system reports exactly what is wrong. This prevents impossible Japanese from entering the system.

The Multi-Language Frontier

The pronunciation system actively processes names through five language cascades: English, French, Italian, Spanish, and Portuguese. The calligraphy engine already supports 13 languages internally — the five active ones plus German, Dutch, Swedish, Finnish, Norwegian, Polish, Russian, and Japanese. The gap is not in the engine's ability to handle these languages, but in the pronunciation dictionaries and conversion rules needed to feed it.

Each language added to the active system improves thousands of pronunciations in a single update. The dictionary data already exists. The work is in building the conversion rules and validating the results — a one-time investment per language that pays off across every name in that language.

Not all languages carry equal weight. German and Dutch names are among the most common in the uncovered inventory. Swedish and Finnish are smaller but important for Nordic customers. Polish and Russian serve large diaspora populations. The English bridge means no customer is left without any pronunciation, but the quality difference between an English approximation and a native-language pronunciation is meaningful — especially for languages where English approximations are poor.