About MieruTone

A cozy pitch accent trainer built on quality data and transparent methodology.

MieruTone ("visible tone") helps Japanese learners see and hear the melody of the language. We combine trusted linguistic databases with intuitive visualization to make pitch accent accessible to everyone.

Data Sources

Kanjium Pitch Accent Dictionary

323,000+ entries

Our primary source for pitch accent patterns. Kanjium is a comprehensive open-source dictionary derived from authoritative Japanese sources including NHK and other accent dictionaries.

Primary
github.com/mifunetoshiro/kanjium

UniDic (NINJAL)

Cross-validation

We use UniDic from Japan's National Institute for Japanese Language and Linguistics to cross-validate accent patterns and provide fallback data when Kanjium doesn't have an entry.

Validation
clrd.ninjal.ac.jp/unidic

JMdict / EDICT

264,000+ entries

English meanings and parts of speech come from JMdict, the most comprehensive freely available Japanese-English dictionary, maintained by the EDRDG.

Meanings
edrdg.org/jmdict

Methodology

Confidence Indicators

Not all pitch data is equally reliable. We show confidence levels so you know when to trust the pattern:

Solid dots - High confidence (dictionary match)
Dashed dots - Particles (follow context)
?Question mark - Unknown/uncertain
UniDicBadge - Fallback source used

Compound Word Analysis (McCawley Rules)

When compound words aren't in the dictionary, we predict accent using McCawley's rules based on the second component's mora count:

N2 LengthRuleExample
1-2 moraeAccent on last mora of N1日本語 (nihongo)
3-4 moraeAccent on first mora of N2携帯電話 (keitai denwa)
5+ moraeFollow N2's original accent(offset by N1 length)

Dictionary entries always take priority over rule-based predictions.

Data Source Priority

  1. Exact surface+reading match in Kanjium
  2. Lemma match in Kanjium (e.g., 食べた → 食べる)
  3. Reading-only match in Kanjium
  4. UniDic cross-validation / fallback
  5. McCawley rules for compounds
  6. Standard pitch rules (heiban default)

Updates

Jan 2025

Database expansion - Updated to 323k+ pitch patterns with improved compound word detection and confidence scoring.

Dec 2024

Launch - MieruTone goes live with core analyzer, TTS playback, and spaced repetition learning system.

Feedback & Contributions

Found an incorrect pitch pattern? Have suggestions for improvement? We welcome contributions and feedback!

Try the Analyzer

MieruTone is built with data from open-source projects. We are grateful to the maintainers and contributors of these resources.

Pitch accent data follows Tokyo dialect (標準語) conventions.