Windows · Standard and Pro

How to use Drum

Drum is a desktop close-reading tool for authentic foreign-language texts. Tap a word, mark it by stage, and turn what you want to read into your own textbook. Standard is free with no expiry. Pro adds PDF, YouTube, Text Analyzer & Recommendation, and mobile access (same-LAN QR) after you verify a license on this device.

1. Install

There are two installers: Standard (free, no expiry) and Pro (paid; verify a license in the app). See section 10 for the feature split.

1. Click Download on the site for the Standard installer, about 98 MB. The file is still named Drum-Demo-1.0.0-x64-setup.exe (legacy filename; it is Standard). For Pro, see pricing.
2. Run it. You need 64-bit Windows 10 or 11. If WebView2 is missing, the installer will prompt to fetch it.
3. Open Drum from the desktop or Start menu.

Windows SmartScreen may say “Windows protected your PC.” That is common for an unsigned build: More info → Run anyway. You do not need to install Python.

2. First launch

The first start loads a short tutorial plus a few languages and sample texts. The home banner lets you:

• Open tutorial — jump into a sample reading page and try tapping words.
• Clear demo database — wipe sample books and terms so the shelf is empty. This removes everything currently in the database.
• Dismiss — keep the samples, hide the banner.

Change the UI language under Settings → General (English / 中文).

If the home page says you have no languages yet: Settings → Languages, then load a predefined language (English, Japanese, Spanish, …). Japanese tokenization is bundled. Space-delimited languages work out of the box.

3. Import a book

Menu Books → Create new Book.

1. Enter a title and pick the language of the text (the language you are learning, not the UI language).
2. Import source Text to paste a chapter, or File for .txt / .epub (Pro can also take PDF; see PDF).
3. Local audio/video uses the Media tab; see local media. YouTube is Pro-only; see YouTube.
4. For a web article: Books → Import web page, paste the URL. The grab is crude—edit out nav and footers in the book form before you save.

For long texts, a line that is only --- becomes a page break.

Local audio and video (included in Standard)

On Create Book, open Media and pick a local audio or video file. You can paste a transcript, or upload SRT, VTT, or ASS. If you provide no text, Drum transcribes with local Whisper: download a model first under Settings → Speech recognition.

While you read, the text highlights in sync with playback. Pro YouTube import is a separate tab; do not mix it with a local file.

4. Close reading

Open a title from the shelf. Text is on the left; the term form and dictionaries are on the right. The left menu also has focus mode, reading font, bookmarks, and the term list for this page.

• Click a word to open the term card and dictionaries. This does not run AI translate.
• Click and drag across words: two or more words become a phrase term.
• AI drag-translate needs more: select at least three words (API key required). One or two words will not show a gloss.

The term form (the card on the right after you click a word)

The heading is Linguistic Journal. You do not have to fill every field: marking STAGE and hitting COMMIT (or Ctrl+Enter) is enough to save it to your lexicon.

• Headphones: speaks the current word using this language’s TTS code and a system/browser voice.
• The word (large brown type): click to edit spelling. If you drag across several words, this becomes the whole phrase.
• REF.: the term’s id. Saved terms show a number; unsaved ones show NEW.
• Square on the right: optional picture. If a reading-page dictionary tab is an image search, picking a picture fills this box. Click the image, then Delete/Backspace to remove it, and COMMIT.
• ROOTS: parent / lemma. Link inflected forms to the base, e.g. went → go. + ADD adds more parents. Inherit (below) is only available when there is exactly one parent.
• PRONUNCIATION (hidden for some languages): pinyin, kana, IPA—you type it. Turn on “Show Pronunciation field” in the language settings. Japanese reading furigana is under Settings → Japanese, not here.
• MEANING: your gloss or notes. Dictionaries do not auto-fill this; look at the MDX/web tabs, then copy what you want.
• INHERIT: enabled only with exactly one root. The child’s STAGE then follows the parent—useful for irregular forms and derivatives.
• CLASSIFICATION: tags, comma-separated, e.g. literature, verb. Used for filtering on the term list; it does not change tokenization.
• STAGE: learning status (next subsection). Click here or use 1–5 / W / I.
• DISCARD: for a saved term this deletes the term (confirm dialog; children lose this parent). For a brand-new unsaved term it just abandons the edit. To keep the term, close the form or skip COMMIT.
• COMMIT: saves the word, meaning, tags, status, and image.

If Settings → Behaviour has “Auto-add context sentence” on, the current sentence is stored in Library on the other side of the form. + New adds another example.

Term status

Color shows how well you know the word. You can mark status before you write a definition.

• Unknown — not handled yet (default highlight).
• New 1–2, Learning 3–5 — higher number means more familiar. Keys 1 … 5.
• Well Known (W) — you know it; usually no highlight.
• Ignored (I) — names, particles you do not want to study.

Mark as read (word counts)

Statistics do not increase when you turn the page. After you finish a page, tap Complete / Mark as Read at the bottom. Only then does that page’s word count land in today / this week / this month. The same action indexes this page’s sentences for the term form. If you skip it, Sentences is usually empty. See No example sentences on a term.

Next Step / Mark Rest Known sets leftover unknown words to Well Known, marks the page read, and goes to the next page. When the book is done you will see “Reading Completed.”

Shortcuts

The full list is in the app: Settings → Keyboard shortcuts. Under Settings → Behaviour, turn on “Stop audio on term form open” and “Auto-add context sentence.”

Hamilton word glosses (included in Standard)

This is not the same as whole-page AI interlinear (Ctrl+Shift+B). That shortcut adds a sentence translation. Hamilton adds a gloss on each word and does not use your API key.

Go to Settings → Term Popups and enable “Show word glosses in interlinear translation.” Then open a reading page.

How long a selection must be

A “word” is one clickable token on the reading page (space-separated in English, tokenizer pieces in Japanese). If the selection is too short, that feature does nothing—it is not broken.

5. Add an MDX dictionary

MDX is the MDict local-dictionary format. Lookups appear in the right pane, offline, and are usually richer than the bundled online dictionaries. Use .mdx files you already own.

1. Keep the file in a stable folder, e.g. C:\Users\YourName\Documents\dicts\. If there is a matching .mdd (images, audio), put it in the same folder with the same name. Drum picks it up automatically; you do not add the .mdd separately.
2. In Drum: Settings → Languages, open the language you are learning.
3. Click ＋ Add dictionary.
4. Use for: Terms. Type: MDX Dictionary.
5. Paste the full file path, e.g. C:\Users\YourName\Documents\dicts\oxford.mdx. A folder path is not enough.
6. Leave the checkmark on so the row stays active. Predefined languages already have a Sentences dictionary—keep it. Saving a language requires at least one Terms dict and one Sentences dict.
7. Save. The test icon on the row can probe a sample word.

Back on the reading page, click a word: an MDX tab appears on the right. The first lookup builds an index; later lookups are fast.

The path must exist and end with .mdx, or save will fail. You can add several dictionaries and switch tabs.

6. What each language setting does

Go to Settings → Languages and open a language. Preset languages rarely need a different parser; you will mostly change dictionaries and TTS. Fields from top to bottom:

Name

Shown on the shelf and in filters, e.g. English. Changing it does not break existing books, but keep it recognizable.

Dictionaries

Each row is one dictionary. Drag the six dots on the left to reorder (higher rows appear first as tabs on the reading page).

• First dropdown: Terms looks up a clicked word; Sentences feeds the leftover “Translate sentence” web dictionaries. For daily lookup, add Terms.
• Second dropdown: how it is shown—see “Embedded vs pop-up” below.
• Middle field: for web dictionaries, a URL with [LUTE] (replaced by the current word); for MDX, the full file path.
• Row icons: checkmark enables/disables; external-link tests the dictionary; trash deletes the row.

Saving requires at least one active Terms dictionary and one Sentences dictionary. Preset languages already include a Sentences row—do not delete the last one after adding MDX.

Embedded vs pop-up: some sites must use a pop-up

Embedded loads the web dictionary in the right-hand pane so you can stay on the page. That only works if the site allows being framed. Many dictionaries (Collins, DeepL, Reverso, and similar) block embedding; the pane stays blank, spins, or says it cannot be displayed.

If that happens, set the row type to Pop-up window. Clicking a word then opens that site in a separate window. This is expected, not a broken setting. In the English preset, Simple Wiktionary is embedded; Collins / Reverso / DeepL are already pop-ups for this reason.

How to check: click the row’s test icon. The test page includes a preview—if you see the dictionary, embedded is fine; if it is empty or refused, switch to pop-up.

• MDX Dictionary always stays in the right pane and is not affected by anti-embed headers. It is still the best daily lookup.
• The desktop window may block pop-ups. Allow them the first time, or turn on Settings → Behaviour → Open popup in new tab. If it still will not open, use MDX, or open http://127.0.0.1:5001 in a browser while Drum is running.

Show Pronunciation field

When on, the term card shows a pronunciation box (pinyin, kana, IPA—you fill it in). Japanese reading furigana is under Settings → Japanese, not this checkbox.

Right-to-left

Turn on only for languages written right-to-left, such as Arabic or Hebrew. Leave it off for English, Japanese, and Russian.

Parse as

How the book is split into clickable words. The wrong parser makes the page unclickable or over-split.

• Space Delimited: English, French, Spanish, Russian, and most other spaced languages. Most presets in Standard use this.
• Japanese: Japanese with the bundled Sudachi tokenizer. Required for Japanese—do not use space splitting.
• Classical Chinese: Literary Chinese, one character at a time.
• Turkish: Turkish, including İ / ı casing.

Modern Chinese, Thai, and Khmer need extra tokenizer plugins. The current installer does not ship them, so they usually do not appear in the list.

Character substitutions

Replacements run on import before tokenization. Format: old=new, several pairs joined with |. Typical use: normalize curly quotes and ellipses. English already has a preset string; leave it unless you know you need a change.

Split sentences at

Characters that end a sentence, default like .!?. This sets how long “one sentence” is for AI drag-translate, interlinear, and copy-sentence. Japanese and Chinese presets also include 。！？.

Split sentence exceptions

Abbreviations that should not end a sentence, joined with |, e.g. Mr.|Mrs.|Dr.|U.S.. Without this, Dr. Smith splits after Dr.

Word characters

Which characters belong inside a word. English is typically a letter range; hyphens may be included here too. A wrong value splits words on click or treats punctuation as part of the word. Keep the preset unless you know the regex.

TTS Language Code

BCP 47 code used to match a system or browser voice, e.g. en-US, ja-JP, ru-RU. A wrong code picks the wrong voice or none at all.

7. Set up a cheap AI API and try translation

Drum does not sell cloud credits. Translation and explain go from your PC to the provider you configure. Preferred: drag a sentence for a gloss; open Explain on the right only when you want grammar. Ignore the leftover “Translate sentence” menu.

Recommended: DeepSeek (inexpensive, works well from China)

Official pay-as-you-go is far cheaper than OpenAI. A small top-up usually lasts a long time. Direct access from mainland China is typical; no proxy required.

1. Open platform.deepseek.com, sign up, create an API key, and copy it (shown once).
2. Add credit on that site.
3. In Drum, Settings → AI:
   • Enable AI translation and explanation
   • Type: DeepSeek
   • Paste the API key
   • Model: deepseek-v4-flash (current official name; deepseek-chat is retired). Use deepseek-v4-pro if you want the stronger, more expensive model.
   • API URL can stay empty / default https://api.deepseek.com/v1/chat/completions
   • Proxy strategy: Direct only
4. Click Test AI Connection, then Save.

Other cheap or free options

• Gemini: Google has a free tier. Type Gemini, model e.g. gemini-2.5-flash. From China you usually need a proxy on the same page, e.g. http://127.0.0.1:7890, strategy Proxy only or Auto.
• OpenAI-compatible relays: Type Custom API, paste their .../v1/chat/completions URL, key, and model name, then test.
• Ollama: If you already run a local model, type Ollama, default http://127.0.0.1:11434/v1/chat/completions. No cloud.

Drag-translate and explain default to Chinese. To change the output language, or to give the model the surrounding sentence, edit the two templates under Settings → Prompts, then save at the bottom of the page.

Prompt templates (including {context})

The shipped templates only include {text}, as in the screenshot. Add {context} yourself so the model can see the whole sentence; that helps with pronouns and word sense.

• {text}: the span you selected, or the current sentence when explaining. Keep it; without it Drum cannot insert the source.
• {context}: the full sentence around that span. If there is no surrounding sentence it becomes empty; the template still works.

Do not add other brace placeholders (e.g. {word}) or the request will fail. Whole-page interlinear (Ctrl+Shift+B) uses a built-in prompt, not these two fields.

Complete templates with {context} are below—paste them as-is. To target English or Japanese, change Chinese to English / Japanese in both.

Translation Prompt Template (drag-select)

Translate the following text to Chinese. Use the surrounding context only to resolve ambiguity (pronouns, word sense). Return only the translated text, with no quotes or explanations.

Text to translate:
{text}

Surrounding context:
{context}

Explanation Prompt Template (AI explain in the right pane)

Explain the following text in Chinese, including grammar, vocabulary and usage. Be concise. Use the surrounding context to resolve ambiguity.

Use a hierarchical list: one top-level numbered item (1. 2. 3. ...) per main phrase or clause; for sub-points (e.g. grammar, word meanings under that phrase) use a nested list—indent with 4 spaces and use "1." "2." or "-" for sub-items. Do not use one flat numbered list for everything.

Text to explain:
{text}

Surrounding context:
{context}

Try it on a reading page

1. Open a book. Drag across at least three words (not just one or two). A translation appears on the selection. That is everyday AI translate.
2. Whole-page gloss: Ctrl+Shift+B. Each sentence gets a line of translation. Press again to hide. That is sentence-level AI. For word-by-word glosses, use Hamilton.
3. AI explain in the right pane auto-runs only when the term text is longer than 20 characters (a short word will not). Standard and Pro both use your own key; Drum does not add a daily cap.

A single click, or a one- or two-word drag, will not show an AI gloss. Drag a sentence, or see “How long a selection must be” above.

8. Read aloud (TTS)

The reading menu item Read Aloud (TTS) uses window.speechSynthesis: the voices available in the current window.

• Desktop window: only voices installed on Windows. If you have not added a Japanese or Russian voice pack, the list is empty or falls back to another language. Add voices in Windows Settings → Time & language → Speech.
• Open the same Drum in a browser: you get that browser’s voices. Edge usually has a fuller set (including Microsoft neural voices).

To open Drum in a browser:

1. Start Drum as usual and leave it running.
2. Open Edge (or Chrome).
3. Go to http://127.0.0.1:5001. This is local; it does not go out to the internet.
4. You see the same shelf and terms. Read-aloud here uses the browser’s TTS.

Drum must stay running. If you quit the desktop app, that address will not load.

9. Terms and stats

Saved words appear under Terms. Filter by language and status. Existing lists can be bulk imported. To make review cards, use Anki export.

Practice → Type cloze-deletes the target word in its original sentence. Listen with TTS, then type it.

Reading volume lives under About → Statistics, per language. Again: only pages you marked as read are counted.

Bulk import terms

Menu Terms → Import Terms. Use a CSV to add a vocabulary list, or overwrite translation, tags, and status on terms you already have.

1. Use a UTF-8 CSV. The first line must be a header with at least language and term. Column order does not matter.
2. Choose the file. Tick “add terms that are not already in the list” and/or “overwrite existing terms.” To pre-fill a meaning but still show the word as unknown while reading, also tick “new terms as Unknown.”
3. Import. The language value must match the language name saved in Drum exactly (for example English).

Optional columns: translation, parent (lemma / infinitive), status (1–5, W Well Known, I Ignored; default 1), tags (comma-separated; quote the cell if a tag contains a comma), pronunciation, link_status (Y with a single parent links STAGE to that parent). The added column is ignored. Any other heading raises an error.

A field may contain a line break if the whole cell is quoted. Multiple parents: "sentirse, sentarse". Spreadsheet export (Excel / Google Sheets → CSV) is the safest way to build the file.

Example

language,term,translation,parent,status,tags,pronunciation
Spanish,gato,cat,,W,"animal, noun",GA-toh
Spanish,gatos,cats,gato,1,,
Spanish,se sienta,sits down,"sentirse, sentarse",1,,

Row 1 creates Well Known gato. Row 2 creates the plural and parents it to gato (the parent is created if missing). Row 3 attaches two parents to a phrase.

Export to Anki

Drum does not run spaced repetition itself. It sends terms to Anki on this computer. You need Anki running, the AnkiConnect add-on, and local connections allowed.

1. Settings → Anki: enable export. The URL defaults to http://127.0.0.1:8765. Tap “Test AnkiConnect connection,” then save.
2. In Anki, create the target deck and note type first (built-in Basic is fine).
3. Settings → Anki exports → create an export. Name it, pick the deck and note type, then map fields.

A mapping puts a Drum placeholder into an Anki field. Extra text around a placeholder is allowed, e.g. ___ { term }. Common placeholders:

Leave criteria blank to export every matching term. Or write:

• language:"English" — must match the name in Drum
• status<=3 — numeric status: 1–5 learning, Well Known is 99, Ignored is 98
• tags:"verb" or tags:["der", "die"]
• parents.count=1, has:image
• Combine with and / or, e.g. language:"German" and status<=3

Anki requires the first field of a note to be non-empty and unique. Use a throwaway deck while you test mappings. If you rename a deck or field in Anki, update the export in Drum. You can export from the term list or the reading page once AnkiConnect is on.

10. Standard vs Pro

These are two installers, not a switch inside one app. Standard is free with no expiry. Pro needs a license, then Verification on this device before Pro features unlock.

Your books and terms stay on this computer. Drum does not sell cloud AI quota; both editions use your own key.

Pro: device verification

After installing Pro, open Verification, enter your license key, and tap Verify and unlock. This device can then use PDF, YouTube, Text Analyzer, and Mobile access. Until then those entries stay locked; local media and Hamilton still work.

The Standard installer does not include those Pro entries. A license key cannot turn Standard into Pro—install the Pro build.

PDF close reading and OCR (Pro)

Create Book → File, pick a .pdf. A text layer is extracted when present; scans or image-only PDFs need OCR.

Under Settings → PDF OCR, pick an engine (Google Lens or Windows OneOCR). After import, tap words, mark stages, and Mark as Read as usual.

YouTube close reading (Pro)

Create Book → YouTube, paste the video URL. Public captions can be fetched automatically; if that fails, upload SRT / VTT or paste a timestamped transcript. Auto-fetch sometimes needs a Netscape cookies.txt (optional upload on the same page).

You can polish captions with AI (your own key). Playback highlights the text, similar to local media.

Text Analyzer & Recommendation (Pro)

Menu Books → Text Analyzer & Recommendation. Paste a passage or pick a book already on the shelf. Drum scores unknown-word load against your lexicon so you can judge difficulty before importing a whole work.

Mobile access (Pro)

Menu Mobile access, turn the switch on. Scan the QR code or copy the link. Phone and PC must be on the same private network (Wi-Fi or Ethernet). Turning the switch off ends paired sessions. The switch can only be changed from the desktop app.

11. FAQ

Word counts stay at 0

Tap Mark as Read at the bottom after you finish a page. Turning the page or closing the window does not record stats.

No example sentences on a term

Drum only collects example sentences from pages you have marked as read. If the word clearly appears in a book but Sentences / Library is empty, that page probably has not been marked yet. Mark it, then open sentences again.

You can also add a line with + New in the term form’s Library. Word counts and sentence indexing share the same switch: turning the page does neither.

No sound in the desktop window

Install a Windows voice pack for that language, or keep Drum running and open http://127.0.0.1:5001 in Edge to use browser voices.

AI translation does not appear

Test the connection in Settings first. On the reading page, drag at least three words—a click or a two-word drag will not translate. Whole-page gloss: Ctrl+Shift+B. See How long a selection must be.

Cannot save a language / MDX path rejected

Each language needs at least one active Terms dictionary and one Sentences dictionary. The MDX path must be a real file ending in .mdx.

“Translate sentence” in the menu does nothing

That is leftover. Use MDX + AI drag-translate instead. If you still want a web dictionary popup, enable “Open popup in a new tab” under Settings → Behaviour.

The app will not start

Confirm 64-bit Windows 10/11. If a dialog asks for the Visual C++ runtime, follow that link, then retry. Still stuck? Post on the community board or email f12138752@gmail.com.

YouTube / PDF does nothing

Standard does not include those two. Local audio/video is in Standard (Create Book → Media). PDF and YouTube need Pro, unlocked on the Verification page. On Standard, convert the material to TXT, EPUB, or paste.

No predefined Mandarin Chinese

The current installer does not bundle the modern-Chinese tokenizer plugin. Japanese, English, and other space-delimited languages work. Classical Chinese is split per character and generally still usable.

Dictionaries are empty

Open Settings → Languages and add an MDX path as above. Bundled online dictionaries need a network; MDX does not.

Where is my data? How do I back up?

Books, terms, and settings live here:

C:\Users\YourName\AppData\Local\Drum

Use Backup in the app for a zip (manual or daily). Copy that file to another PC to restore. Uninstall may leave this folder; delete it if you want a clean removal.

More questions

Post on the community board (no account) or email us. Pro pricing is on the homepage.