Convert Japanese bank statements to Excel
Tallyscope reads Japanese column headers, dates and numbers in your browser and checks every row against the running balance before you export to Excel or CSV.
What Tallyscope reads in Japanese (日本語) statements. The site is in English; no bank names are used or implied.
Japanese bank statements to Excel
Japanese statements mix scripts, put the year first and sometimes use an imperial era instead of a Gregorian year. The file is read in your browser tab and every row is checked against the printed running balance. The Japanese-specific parts follow.
Japanese column headers the reader recognises
| Column | Japanese words |
|---|---|
| Date | 日付, 取引日 |
| Description | 摘要 |
| Debit (money out) | 出金, お支払金額 |
| Credit (money in) | 入金, お預り金額 |
| Balance | 残高 |
The statement-level label 前月繰越 is recognised as an opening balance. Matching ignores case, accents and punctuation, and the lists hold more words than shown. An unrecognised header can be assigned by hand in the column mapper.
A Japanese sample, row by row
The figures below are synthetic and use Japanese conventions for headers, dates and numbers. The opening label 前月繰越 is the statement's starting balance. The same table, saved as a CSV, is read by the real parser in our automated tests, so what follows is what the reader actually does with it.
| 日付 | 摘要 | 出金 | 入金 | 残高 |
|---|---|---|---|---|
| 前月繰越 | 145,870 | |||
| 2026年1月2日 | 給与 | 290,700 | 436,570 | |
| 2026年1月15日 | 家賃 | 88,300 | 348,270 | |
| 2026年1月20日 | 食料品 | 10,861 | 337,409 | |
| 2026年1月28日 | 返金 | 2,307 | 339,716 | |
- 2026年1月2日 is read as 2 January 2026. 290,700 in the 入金 column becomes +290,700, and the printed balance 436,570 is 436,570, which equals 145,870 plus 290,700.
- 2026年1月15日 is read as 15 January 2026. 88,300 in the 出金 column becomes -88,300, and the printed balance 348,270 is 348,270.
- 2026年1月20日 is read as 20 January 2026. 10,861 in the 出金 column becomes -10,861, and the printed balance 337,409 is 337,409.
- 2026年1月28日 is read as 28 January 2026. 2,307 in the 入金 column becomes +2,307, and the printed balance 339,716 is 339,716.
Opening 145,870, plus the four movements (net +193,846), gives 339,716, the last printed balance 339,716. Every row agrees with the running balance, so this statement is shown as Verified.
Money out and money in in Japanese statements
This sample uses two money columns, a layout found in many Japanese-language statements: 出金 holds money out and 入金 holds money in, and each amount is printed without a sign because the column carries the direction. The reader takes the direction from the column and then checks it against the balance; if a column were swapped, the balance would not follow and the statement would not be marked Verified.
Japanese OCR at a glance
| Model code | jpn |
| Writing system | Han characters and kana |
| Bundled English model | Cannot read Japanese letters; a scanned Japanese statement needs the jpn model |
| Text PDF, spreadsheet, bank file | No OCR and no model download |
Dates in Japanese statements
The forms 2026年1月15日, 令和8年1月15日 and R8.1.15 all come out as 15 January 2026. Era years are converted from the era base: Reiwa (令和, or R), Heisei (平成, or H) and Showa (昭和) are recognised.
Examples that are read correctly: 2026年1月15日, 令和8年1月15日, R8.1.15.
Numbers and signs
The yen sign is mapped to the currency code JPY. A yen amount like ¥1,234 has no decimal part; since a comma could be a thousands group or a decimal, such amounts are marked as ambiguous and resolved from the rest of the document. Full-width digits and punctuation are normalised to their ordinary forms before parsing.
Examples that are read correctly: ¥1,234. The running balance is the final judge of every sign.
Scanned Japanese statements need the Japanese OCR model
Japanese text needs a model that knows kana and kanji. The bundled English model does not. The app lists a Japanese model (jpn) and recognises the writing system of a read to decide whether to try it. With the OCR language set to English only, nothing is downloaded; in automatic mode (the default) a poor read can trigger up to three trial model downloads from the jsDelivr CDN (cdn.jsdelivr.net), plus any language remembered from this session or detected earlier, and no statement content is ever sent.
The file stays in your tab (every network use is listed here).
What is specific to Japanese
Column titles are recognised without spaces: unspaced scripts are matched by looking for a known header word inside a longer cell, so a title with a unit or a note after the word still resolves. Examples: 日付 as date, 摘要 as description, 出金 and 入金 as withdrawal and deposit, 残高 as balance.
The labels お支払金額 (payment amount) and お預り金額 (deposit amount) are read as money out and money in. The opening label 前月繰越 is recognised; for the closing figure, check the statement status after loading rather than assuming it.
Convert a Japanese statement
Add the file to Tallyscope (10 files at once are free), read each statement's status, fix flagged rows on the Review tab, then export: CSV is free for the first 25 rows; unlimited CSV and Excel, PDF, JSON, OFX, QuickBooks CSV, Xero CSV and a reconciliation report (PDF, plus an Excel sheet) come with the Day pass or Pro (pricing). Several Japanese and non-Japanese files can be merged into one table.
No accuracy percentage is claimed: layouts vary, and the arithmetic check says whether a given statement reconciled (how verification works).
Frequently asked questions
Does Tallyscope read Japanese bank statements?
It recognises Japanese column headers, balance labels, dates and number formats, and checks the rows against the running balance. The website itself is only in English.
Do I need to download an OCR model for Japanese?
For scanned images, yes: Japanese script cannot be read by the bundled English model, so a Japanese model (jpn) is downloaded from the jsDelivr CDN (cdn.jsdelivr.net) when you choose that language or auto-detect needs it. Text PDFs and spreadsheets need no OCR.
Are my Japanese statements uploaded?
No. Files are read in your browser tab. The only OCR download is a language model file from the jsDelivr CDN, requested only when a scan needs a non-English model; no statement content is sent with it.