PhraseTrainer Manual

1. What is PhraseTrainer

A pronunciation training app that lets you clip just the part you want from any material — movies, dramas, podcasts, meeting recordings — listen to it repeatedly, and record and AI-analyze your own pronunciation.

How it differs from conventional pronunciation apps:

• No fixed teaching material. Practice with the phrases you actually want to say (BYOC: Bring Your Own Content).
• Every phrase you practice is kept permanently — re-practice or export anytime.
• Works on both PC and smartphone.
• Separate background music and ambient sound to focus on the dialogue (Voice extraction).
• Japanese / English UI (German and French planned).

2. Basic flow

Step 1: SETUP (PC recommended)
  Drop a video/audio file -> select a range on the waveform -> register the phrase

Step 2: PRACTICE (PC / smartphone)
  Replay the original -> record -> the AI-recognized speech shows instantly
  -> side-by-side comparison with the original -> evaluate & AI commentary as needed

Step 3: HISTORY (PC / smartphone)
  Pick a past phrase -> re-practice -> compare scores
  -> track progress with study status (New/Learning/Done)
  -> organize phrases with Active / Library

You can prepare material on a PC and practice on a smartphone.

3. Setup — Clipping phrases

Drag and drop a video or audio file and clip the section you want to practice.

• Fast even with large files — processed inside your browser; only the needed section is uploaded.
• Set the start and end points (A-B points) intuitively while watching the waveform.
• The "Clip & Upload" button is enabled only once you have selected a range (drag on the waveform, or double-click a subtitle row to set A-B). While no range is selected the button is disabled, preventing an unintended section from being uploaded by mistake.
• If an SRT (subtitle) file is available, it loads automatically and you can clip line by line.
• For videos with multiple audio tracks (e.g. dual-language), the English audio track is selected automatically. The waveform, preview playback and the clip are all in English (single-track videos behave as before; no language picker is needed). On a multi-track video the picture keeps playing while only the sound follows the English track — even when Japanese is the file's default track, English plays.
• You can tag clipped sections (two tag systems: title tags / phrase tags).

Library — bulk import from a folder

The "Open folder" button in the "Library" panel on the left lets you load material in bulk, by folder.

• Register multiple video/audio files at once from within a folder.
• If an SRT subtitle file is in the same folder, it loads together with the matching video/audio (match the file names to ensure reliable pairing).
• If there are subfolders, items are grouped by subfolder.

Instead of dragging and dropping one file at a time, you can import a whole drama season or dozens of podcast episodes at once.

Automatic range selection from SRT subtitles

When you load an SRT subtitle file, a list of subtitles (lines) appears on the Setup tab.

• Double-click a subtitle row to automatically set the A-B points (start/end) to that subtitle's time range.
• No need to drag the waveform manually — clip line by line, accurately and quickly.
• Subtitle rows can also be tagged, so you can quickly find a target line with search and filters.

Clip multiple lines at once with checkboxes

Besides double-clicking one row at a time, you can select multiple subtitles and clip them all at once. This makes it easy to register the lines you want to practice from a single news clip or drama episode in one go.

• Use each row's checkbox to select multiple lines you want to practice.
• Click "Clip selected" to clip up to 10 lines at once and save them to History in a single batch.
• The number you can batch-clip is min(10, your remaining plan quota). When you are near the Active phrase limit, only the remaining quota is clipped.
• Lines longer than 30 seconds are skipped automatically (each phrase is 30 seconds max).
• After batch-clipping you stay on the Setup tab and the result is shown in a banner (it does not jump to the practice screen).
• While clipping, a progress bar ("Clipping… 3 / 10") covers the subtitle list and the list cannot be operated until it finishes. Keep this page open and wait (leaving the page mid-way interrupts the clipping).
• When the progress bar disappears and the result banner appears, clipping is done. Audio analysis then continues automatically on the server, so you can safely leave this page afterward.

Editing the interval of a clipped phrase

When you want to change the interval (start / end) of a clipped phrase, you can overwrite it in place instead of re-creating it.

• Press the "✎" (edit) button at the right end of a "✓ Done" row. That phrase's interval loads into A–B and you enter edit mode.
• Adjust A–B (start / end) and press "Update". The same single phrase is overwritten with the new audio and interval (no new phrase is added).
• The update is reflected immediately in Practice, History and Dictation. Only the latest version is kept — old data is not retained (Export also outputs the latest).
• Press "Cancel" to abort. The same steps also let you re-record just the audio without changing the interval.

Changing and clearing the loaded subtitles

Once an SRT subtitle is loaded, you can switch it for another file or delete it completely.

• "Change" button (to the right of the file name): closes the current subtitle list and returns to the drop zone. Use it when you want to load a different .srt / .json file
• "Clear (delete from DB)" (inside the ⋯ menu next to "Save"): discards the current subtitles and also deletes the saved data. After you confirm, it returns to the drop zone and the subtitles will not be restored even if you reopen the same video

Paging for long subtitles

For videos with very many subtitles (over 200 lines), the subtitle list is shown in pages of 200 rows. Use ◀ ▶ below the list to move between pages. This prevents waveform Zoom and range dragging from becoming heavy on videos with many subtitles. Search works across all pages, and when you double-click a subtitle row to reflect it onto the waveform, the view jumps automatically to that row's page. Subtitles of 200 lines or fewer are shown in full as before.

Editing subtitle rows (merge / split / delete)

Loaded subtitles can be tidied up row by row. Action icons sit at the right edge of each subtitle row. Even when the subtitle breaks are awkward, you can reshape them into easy-to-practice units before clipping.

• Merge (up-arrow icon): combines the row with the row directly above it. The time becomes "start of the upper row → end of this row," and the text and tags are joined. Use it to turn one sentence split across two lines into one (not shown on the first row, since there is no row above to merge with).
• Split (scissors icon): divides the row into two rows. In the window that opens, click the word boundary where you want to split; the time is apportioned by character count and the tags are carried over to both rows. Handy for breaking a long line into units of 30 seconds or less.
• Delete (trash icon): removes that subtitle row from the list. Use it to drop unwanted lines (sound effects, speaker labels, etc.).

📱 Handling large videos on iPhone / iPad

Safari and Chrome on iPhone / iPad have a per-tab memory limit of about 1–1.5GB. Loading large files (rough threshold: 700MB or more) may cause the browser to crash or unexpectedly switch tabs during loading.

• Recommended workflow: For large videos (a full movie, a single TV episode, etc.), clip phrases on PC first, then practice the resulting phrases on iPhone / iPad
• iPhone / iPad is designed as a practice device; Setup (clipping) on mobile is intended for supplementary use
• If you try to load a file over 700MB on iPhone, a warning dialog blocks the load (no "continue" button — this prevents crash loops)
• Short videos and audio of a few hundred MB are usually fine for clipping on iPhone

4. Practice — Practice, evaluation & AI analysis

Playback

• Repeat playback of the original (1x / 2x / 5x / infinite loop)
• Switch between original / Voice version / noise-reduced version
• Pinpoint-play just a specific word on the waveform (range select, touch supported)
• ▶ / ⏸ toggle button — during playback the button turns into a stop icon; stop with one tap

Recording

• Start / stop with the record button, a hotkey, or a foot pedal
• Designed so the start of your speech isn't cut off, even on consecutive recordings
• Trim unwanted parts of the recording
• Right after you stop recording, the "AI-recognized speech" appears within 1-2 seconds, without pressing the evaluate button

The ANALYSIS panel — two zones

The ANALYSIS panel splits its content into two zones: "model information you can read without recording" and "your own results that change with each recording."

• 📘 Original Analysis (fixed · no recording needed) — difficulty, sound-change rules, translation, and the original words/phonemes are shown the moment you open a phrase. You can see "where this sentence is hard" before you even record.
• 🎤 Your Evaluation (per recording · saved to history) — scores, the side-by-side comparison, and pronunciation advice update every time you record and evaluate, and are saved to your history.

Difficulty badge

Each phrase shows a difficulty badge (Easy / Medium / Hard). It is an automatically computed estimate of "how hard to pronounce" the phrase is, derived from sound changes (linking, elision, weak forms, etc.) and phoneme complexity (an internal score of 0-100, where 100 is hardest). Each History card also shows it as "Lv NN," a handy guide for prioritizing harder phrases.

⏳ Analyzing… and 🔄 Re-analyze

The original's phoneme alignment (which computes the sound-change rules and difficulty) runs on the server the first time you open that phrase in Practice. It can take a few tens of seconds, during which "⏳ Analyzing…" is shown. Wait a moment and the results fill in automatically.

• For phrases where analysis is incomplete or failed, a "🔄 Re-analyze" button appears. Press it to run the analysis again.
• For phrases that can't be analyzed (original too short, silent, etc.), sound-change rules may not be shown.

Comparison & evaluation

• Side-by-side comparison — your speech and the original lined up word by word, with differing words color-highlighted
• IPA notation — phoneme sequences shown in the International Phonetic Alphabet (e.g. h ə l oʊ). Switch to ARPAbet notation (HH AH0 L OW1) with the "IPA ⇄" header in the comparison table.
• Cambridge Dictionary integration — click a word in the original to open the Cambridge Dictionary (US) in a new tab
• Waveform linkage — word boxes on the waveform are colored to match the comparison result
• F0 (pitch) curve — visually compare your intonation with the original

AI pronunciation evaluation scores you automatically on 5 metrics plus a total:

Hover over a metric label to see a description in your selected UI language.

AI commentary & translation

• AI pronunciation advice — identifies "✅ words you pronounced as in the original" and "🎯 words you struggled with (challenges)," and for each challenge word shows specific, per-word practice advice. It explains using natural sound-change terms (linking, assimilation, elision, flap T, weak forms) and is returned in your selected UI language.
• AI translation, part-of-speech breakdown & grammar points — the AI explains the subtitle's English in a 3-part structure: translation + part-of-speech breakdown + grammar points. Generated automatically in the background after a new segment is created, and shown in the "Translation" section of the ANALYSIS panel. The part-of-speech breakdown lists every word in order of appearance and always includes both its part of speech and its meaning (e.g. execute: verb (to carry out)), so there is never any doubt about which words are glossed.
• Sound-change rule detection — liaison, elision, weak forms and flap T are detected and highlighted automatically

Collapsing Analysis items

Each Analysis item — word comparison, translation, sound-change rules, AI commentary — can be collapsed by clicking its section header (▼ / ▶ icon). When there's too much information, collapse the items you don't use for a simpler display. Keeping only the items you use often expanded keeps the screen tidy. (The learning curve cannot be collapsed.)

Focus mode (hide the original)

The "Hide" button at the top right of the Original section hides the word labels on the waveform, the subtitle text, and the original side of the AI recognition result all at once. Your recording side stays visible, so you can do dictation-style learning: pronounce by ear without looking at the text, then check the answer afterward. The setting is saved in your browser.

5. History — History & Active / Library management

All the phrases you've practiced are saved in your history.

• Re-practice anytime with one tap — the original, your recordings and scores are all reproduced
• Check score trends on the learning curve graph
• Search and filter by tag, date or text

Per-media numbering badge #03 (06:32)

Phrases cut from the same video/audio get a numbering badge like #03 (06:32).

• #03 = a sequential number ordered by playback time within that media (e.g. the 3rd phrase that appears in the video)
• (06:32) = the timecode where that phrase starts in the original video
• Even if you later cut extra phrases in between and the numbers shift, the timecode tells you the real position, so you never lose track of which scene a phrase is from

The numbering badge is shown both on the History card and in the Practice header (title + number).

Move to the previous / next phrase (◀ ▶)

The Practice header (to the left of the numbering badge) shows ◀ ▶ buttons. Without going back to History, you can step through phrases from the same title (same video/audio) in numbering order — record, next, record… in a smooth rhythm.

• Navigation only moves between Active phrases of that title (Library phrases are skipped), so you are not asked to activate a phrase every time you press Next.
• You can also move with the ← / → arrow keys (the keys are configurable in Settings and can be assigned to a foot pedal).
• At the first / last phrase the corresponding button is disabled. If a title has only one Active phrase, the buttons are hidden.
• If you have an unevaluated recording, moving asks you to confirm that the recording will be discarded.

Narrowing tags (selecting a title scopes the phrase tags)

As you add more tags the list gets long. When you select a title tag in History, the phrase-tag list automatically narrows to just the tags under that title, so only the relevant tags are shown.

• When there are many phrase tags, use the search box to filter them
• Beyond a certain number, tags are collapsed and can be expanded with "Show all"

Study status (New / Learning / Done)

Each phrase has a study-status badge (on the left of the card) for tracking your own mastery. This is a separate axis from the Active / Library practice-slot management below — it purely reflects "how finished you are with it."

• Each click cycles New → Learning → Done → New …
• Recording for the first time automatically sets it to "Learning" (only from New; Learning/Done are never downgraded by recording)
• "Done" is set manually by you — it never becomes Done automatically
• Use the "Study: All / Learning / Done" filter at the top of History to filter by status. It's on a separate row from the Active / Library filter, and you can combine both

Active and Library

Phrases are managed in two layers: Active and Library. This is the billing / practice-slot axis, independent of the study status above.

• Switch the display with "All / Active / Library" at the top of the History tab
• Switch each phrase with the Active / Library toggle switch on the right of the card (green = Active). Tap the switch to flip between Active ⇄ Library
• An Active remaining-count is shown at the top right
• Click a Library phrase and you'll be asked whether to make it Active and start practicing

🎧 Dictation practice

Turn on the switch located right below the History filters (display / study / tags / search / sort / date) and the list turns into a dictation list. Whatever you've currently filtered down to becomes your practice set (you can play both Active and Library phrases).

• Each row shows two lines — line 1 = the original text, line 2 = the translation (only the translation itself is shown; the part-of-speech breakdown and grammar notes are not included)
• Use the small ▶ button at the right of each row to play that row's original audio (a simple listening-focused dictation — no waveform)
• The "Hide original" / "Hide translation" toggles above the list mask that line (independently). A masked cell becomes "tap to show", and a click reveals just that row — so you can listen → recall → reveal to check yourself
• If a phrase's translation hasn't been generated yet, it is generated automatically when that line is shown (no extra usage; nothing is generated while it stays hidden)
• Click the numbering badge (e.g. #01) or the title on the left of a row to open that phrase directly in Practice (the same flow as the normal History list). When you decide mid-listening that you want to record and practice a phrase, you can jump straight to Practice (tapping the original / translation cells still reveals the answer as before)
• CSV export — the "Export CSV" button above the list writes your current filtered results as text for an external flashcard app. Title tags become courses (# lines) and phrase tags become chapters (## lines), and each card is two lines: translation then original (title-only phrases get just the course line; if you leave out chapters, the receiving app splits them automatically). Copy the text shown in the window and paste it directly, or use "Save file" to download it as CSV. Any phrases whose translation isn't generated yet are generated together before output (no extra usage)

📒 Vocabulary

In the part-of-speech breakdown of the Practice AI analysis, every word now has a checkbox on its left. Check a word you want to remember and PhraseTrainer saves a snapshot of the word, its meaning, and its IPA at that moment (so it stays stable even if the translation later changes). Uncheck to remove it.

Checkboxes appear on phrases whose part-of-speech breakdown has been generated. Phrases with no translation yet, or older phrases, may not show the breakdown — in that case press the "Generate translation" button in the analysis pane to rebuild it, and the checkboxes will appear.

Turn on the "📒 Review vocabulary" switch right below the History filters and the list turns into your vocabulary. Only the words that belong to your current filter (title / tags / date, etc.) are shown. The order also follows the History sort (Numbering order / Newest / Oldest) (for score / difficulty sorts, which are meaningless for individual words, the list falls back to numbering order).

• List view — each word on two lines: word + IPA / meaning. The 🔊 button on each word plays its pronunciation on the spot (using your device's built-in voice). You can also click a word to open it in the Cambridge Dictionary
• Card view (flashcards) — study one card at a time. Tap the card to flip between front (word + IPA) and back (meaning), and use ◀ ▶ to move. On a phone you can also swipe left/right to move between cards (left = next, right = previous; a small tap still flips). You can switch the starting face (word-first or meaning-first) with a button. The "🔀 Shuffle" button shows the cards in random order (card view only; press again to return to the normal order, and each press reshuffles). Keyboard shortcuts work too: ←→ (↑↓) move cards / Enter flips / Space speaks the word (your device's built-in voice)
• CSV export — pressing "Export CSV" opens a window showing the CSV text. Use "Copy" to copy it all and paste it straight into Quizlet / Anki ("Save file" still lets you download it as a CSV file). The content is two columns ("word + IPA", "meaning"), and you can choose the separators (between front and back = Tab / Comma, between cards = New line / Comma / Semicolon; the default is the Quizlet/Anki standard "Tab + New line")

6. Export — Exporting your library

You can export your entire phrase library as a self-contained archive. Permanently yours — your learning assets stay with you even after you cancel.

• Regular: CSV / Excel — outputs phrase text, tags, scores and audio links as CSV and Excel files
• Pro: full HTML archive — a ZIP package that works fully in a browser, including the original audio, Voice version, noise-reduced version, learner recordings and all analysis data. Just open index.html to use waveforms, F0, analysis, the learning curve and audio playback offline.

Run Export from "Library management" inside the ⚙ settings modal.

• Eligibility: after 30 consecutive days of subscription
• Count: up to 2 times per 30 days. Processing is sequential (1 at a time).
• The generated ZIP is downloadable for 7 days

How the archive works

Export (the archive) and the online service (practice, AI evaluation, preview) play separate roles. Together they give you both "your learning assets are yours for life" and "the service is always up to date".

• The archive is a snapshot at the moment you export. Practice you do afterward is not automatically added to an existing ZIP. To capture the latest records, keep practicing online or run Export again.
• There is no Import feature. All of your past phrases are kept permanently in your Library. You can practice and re-evaluate any of them again simply by moving them back to Active, so there is no need to re-import from an archive.
• The archive works fully offline. The original and noise-reduced audio (plus the Voice version, learner recordings, analysis data and HTML viewer for Pro) stay on your device and play back without an internet connection, even after you cancel.
• The latest AI evaluation and re-analysis happen online. Think of the archive as "for taking out and keeping" and the online service as "for practicing, evaluating and improving" — that keeps things clear.

7. Audio processing — Voice extraction & noise reduction

Voice extraction (Pro only)

AI separates the background music and sound effects of movies and dramas to extract just the dialogue. Even with material that has loud BGM, you can practice with clean, easy-to-hear audio. After extraction, you can switch between "original" and "Voice version".

Noise reduction (all plans)

Automatically removes recording-environment noise such as air conditioning, fans and hiss. You can switch between the original and the noise-reduced version, and combining it with Voice extraction makes it even clearer.

8. Settings, hotkeys & foot pedal

Open the settings modal with the ⚙ button at the top right of the toolbar.

• Hotkeys — Space = play, R = record, E = evaluate, etc. Key assignments are customizable
• Theme — light / dark
• Language — switch UI language (Japanese / English)
• Font size — small / medium / large
• Pronunciation comparison — customize the symbols ignored in comparison (. , ! ? ; : etc.)
• Foot pedal — hands-free play/record with a USB foot switch (operable even with both hands full)

9. Plans & pricing

Upgrade from the pricing cards on the landing page, or from the "Upgrade" item in the top-right menu after logging in. You can choose monthly / yearly (20% off) billing, applied instantly via Stripe. Cancel from "Manage subscription"; your current plan stays active until the next billing date.

10. Using on a smartphone

• When you access from a smartphone, the History tab opens automatically
• You can prepare material on a PC and practice on a smartphone
• A smartphone microphone is often higher quality than a PC's, improving recording quality
• Setup tab on mobile: the SRT subtitle list sits right below the waveform / Zoom, so you can pick a range while reading the subtitles. A-B times, Reference and the title tag are tucked into a collapsible "Details" panel above the waveform (open it only when needed), and the clip button stays docked at the bottom of the screen. Tapping a subtitle row moves the waveform range, so you can work even with Details collapsed. The SRT controls (Change / Font size / Marker / Save / Search) are also folded into an "SRT options" panel so the subtitle list gets as much room as possible.
• Full support for touch operations on the waveform (range select, playback, trimming)

11. Known issues (beta)

• Old iPad (iOS 14 or earlier): a click sound may occur on the first playback (it does not happen from the second time on)
• Some versions of Safari: about the first 100ms after recording starts may be cut off
• Voice extraction: with material mixing classical music or loud BGM, parts of the dialogue may become faint
• Monthly recording-limit counter: deleting a phrase does not roll back the recording count (it resets automatically per calendar month)

12. Contact

• Email: [email protected] (we reply within 1 business day as a rule)
• Bug reports: attaching a screenshot and the steps helps us identify the issue faster
• Feature requests: welcome, big or small

Before registering, please also review the Terms of Service / Privacy Policy / Legal notice (Specified Commercial Transactions Act).

13. What's New

A list of the main features and changes recently added to PhraseTrainer (newest first). For how to use each one, see the linked section.