A reading pacer for text files.
SpeedReader.ahk — A speed reading trainer for plain-text files
Author Steve (kunkel321) with Claude (Anthropic)
Version Date (see code)
Forum https://www.autohotkey.com/boards/viewtopic.php?f=83&t=140586
Github https://github.com/kunkel321/SpeedReader
Requires AutoHotkey v2.0+ | RichEdit.ahk by just-me (place in Tools\ folder) https://github.com/AHK-just-me/AHK2_RichEdit
SpeedReader opens a .txt file and highlights a configurable "chunk" of words at a time, advancing through the text at a user-set WPM rate. Pauses are automatically inserted at sentence, paragraph, and list-item boundaries. The goal is to train faster reading by giving the eye a single focal point that moves at a controlled pace. Works well with plain-text books from Project Gutenberg and similar sources.
Portable mode (recommended): rename AutoHotkey64.exe to SpeedReader.exe and place it in the same folder as SpeedReader.ahk and RichEdit.ahk. No installation needed. Standard mode: run SpeedReader.ahk directly with AHK v2 installed on the machine. Either way, RichEdit.ahk must be in the Tools\ folder.
- On first run, srSettings.ini is created automatically with sensible defaults.
- Open a .txt file via File > Open, by dragging a file onto the window, or by choosing from the recent-files list in the File menu.
- Press Play (or the Down arrow key) to begin. Press again to pause.
Play / Pause ▶ Play button —or— Down arrow key (window must be active) Restart ⟲ Restart button — clears highlight and rewinds to word 1 Speed (WPM) Big slider at the bottom of the window, plus: Left / Right arrow keys (±WPMHotkeyStep per press, default ±25) Mouse wheel over the slider (wheel-up = faster, wheel-down = slower) Jump to any word Double-click the word — playback restarts from that point immediately. Works whether the reader is playing, paused, or stopped. Open file File > Open (Ctrl+O), or drag-and-drop a .txt file onto the window Focus search box Ctrl+F (pauses the pacer if running; selects existing text) Recent files Numbered list in the File menu (files that no longer exist are omitted) Resume position When a file is re-opened that has a saved position, a prompt offers to resume from the last reading point or start over. Position is saved automatically on pause and on window close.
▶ Play / ❚❚ Pause Toggle playback ⟲ Restart Rewind to the beginning Size Font size spin-box — live update, no restart needed Chunk Words highlighted per tick (1–10). Larger = easier at high WPM. Status bar Filename + word count while stopped; "Word X of Y (Z%)" while playing
WPM: N [m:ss] Large WPM readout. Once ~10 words have been read a rolling time-remaining estimate appears in [m:ss] format, updated at the interval set by TimeRemainingUpdateMs (default once per second). The estimate uses actual dwell times, so it accounts for sentence pauses, smart pacing, chunk size, etc. automatically.
Overlap Sliding-window mode: the chunk-sized highlight advances one word per tick instead of jumping by a whole chunk. Gives a smoother, flowing feel. Sentence/list pause multipliers are suppressed in this mode (they would disrupt the slide); Smart pacing still works.
Sentence pause Dwell longer at sentence ends (. ! ? …) and strong mid-sentence breaks (; : em-dash). Paragraph ends always pause regardless. Comma pauses are lighter and never clip a chunk boundary. Common abbreviations (Mr. Dr. etc. e.g. p. …) are excluded from sentence detection so they don't trigger false pauses.
List pauses Detect short lines that look like list or enumeration items and pause at the end of each. A line qualifies if it is short (≤ ListMaxWords words), doesn't end in a comma/semicolon/hyphen, doesn't end in a bare conjunction (and/or/but…), and the next line starts with a capital letter.
Smart pacing Scale each word's dwell time by its reading difficulty: Stopwords (the, of, and …) → StopwordWeight × base (faster) Monosyllabic content words → MonoWeight × base (baseline) Polysyllabic words → +ExtraSyllableWeight per extra syllable beyond the first (slower) Syllables are counted with a vowel-group heuristic; silent trailing 'e' is subtracted. Good enough for relative weighting purposes.
Center scroll Keep the highlighted word vertically centered in the reading pane as the text advances. When off, the pane scrolls only when the highlight reaches the bottom edge (standard behavior). Scrolling is in whole-line increments (RichEdit limitation).
Read Aloud Speaks the text using Windows' built-in SAPI5 text-to-speech. SAPI is queued one sentence at a time; within a sentence the highlight follows SAPI's own word-boundary events, so the voice and the highlight stay in genuine sync. If a voice doesn't report word boundaries, an internal fallback pacer (the same WPM/weighted dwell-time math the silent pacer uses) carries the highlight instead, and each sentence's end resyncs to the last word once SAPI actually finishes speaking it. Only available at or below TTSMaxWPM (see the tunables near the top of the script) since spoken word stops being intelligible well before the top of the WPM slider's range; the checkbox disables itself automatically above that speed, and re-enables itself when WPM drops back down. Chunk / Overlap / Sentence pause / List pauses / Smart pacing are all timer-engine concepts and are grayed out while Read Aloud is active. Play/Pause and double-click-to-jump work the same as always. The Voice dropdown next to the checkbox lists every SAPI voice installed on this machine; "(Default voice)" leaves whatever Windows considers the default in place. Changing the voice takes effect on the next launch, not immediately — restart SpeedReader after picking a new one.
Highlight / Text / Background Clickable colored buttons — click to open the Windows color picker. Changes apply immediately to live text. Font Drop-down of preset font faces (condensed). Search Incremental search box — results highlighted as you type (debounced). Literal by default — "first second" matches the exact phrase, including punctuation ("Mr. Darcy" matches literally). Matching is case-insensitive and can span multiple words. Regex mode: prefix the term with '/' (optional trailing '/'): /first\s+second/ — first + any whitespace + second /\bthe\s+\w+ing\b/ — 'the' followed by an -ing word /^Chapter \d+/m — multi-line anchors work Invalid regex patterns show a brief error in the status bar. Ctrl+F focuses the search box (pauses the pacer if running). F3 advances to the next match (honors current mode). Disabled while the pacer is running; re-enabled on pause. Clearing the box removes the search highlight immediately.
File > Open Standard open dialog, filtered to .txt files (Ctrl+O) File > Convert PDF/EPUB/TXT… Launches TextExtractor.ahk companion tool File > Open srSettings.ini Opens the INI in your default text editor File > Recently converted Submenu of *.txt files in the Converted\ subfolder (newest first) File > 1 … N Recent files, most-recent-first (missing files skipped) File > Exit Links > Project Gutenberg https://www.gutenberg.org/ebooks/results/ Links > AutoHotkey forum Forum thread for this script (set URL_AhkForum below) Links > GitHub repo Repository for this script (set URL_GitHub below) Debug > Token analysis ListView of tokenization flags for ~80 words around the current reading position. Columns: index, text, endsSentence, endsParagraph, endsLine, endsListItem, endsCommaLike, weight. Ctrl+C copies all rows as CSV.
SpeedReader polls srSettings.ini every 1.5 s for a new TEConvertedAt timestamp written by TextExtractor after each successful conversion. When a new stamp is detected the converted file is loaded silently; the status bar shows the filename. The polling only fires when the SpeedReader window is active, so there are no stale loads while the user is working in TextExtractor.
All GUI settings save automatically on change and restore on next launch. [Reader] WPM, ChunkSize, Overlap, SentencePause, ListPauses, SmartPacing, CenterScroll [Colors] HighlightColor, TextColor, BackColor (stored as decimal RGB integers) [Font] Name, Size [srWindow] W, H (saved on close) [Session] LastFile, TEConvertedAt LastFile is stored as "path" or "path, wordIdx" and auto-loaded on launch. TEConvertedAt is the sentinel timestamp written by TextExtractor. [Recent] File1 … FileN — each value is "path" or "path, wordIdx". The word index is the saved reading position for that file. Missing files are silently skipped on load.
Pacing object Dwell-time multipliers applied on top of the base WPM rate: SentenceMult after .!?… (default 1.8×) ParagraphMult after blank line (default 2.5×) CommaMult after comma (default 1.15×) SemicolonColonMult after ; : em-dash (default 1.5×) ListItemMult after list item (default 1.4×) StopwordWeight fast-word weight (default 0.5) MonoWeight baseline weight (default 1.0) ExtraSyllableWeight per extra syllable (default 0.3) WPMHotkeyStep Arrow-key WPM step size (default 25) ListMaxWords Max line length in words for list-item detection (default 12) TimeRemainingUpdateMs How often [m:ss] refreshes, in ms (default 1000 = once/second) RecentFilesMax How many recent files to remember in the File menu (default 9) SearchFromCurrent true → search starts from current reading position (default) false → search always starts from top of document SearchDebounceMs Delay (ms) after last keystroke before search fires (default 300) URL_AhkForum AHK forum thread URL — fill in when known URL_GitHub GitHub repo URL — fill in when known
Uses Just Me's RichEdit.ahk wrapper (RICHEDIT50W from Msftedit.dll, RichEdit v4.1).
All character offsets are zero-based UTF-16 code units, matching RichEdit's internal indexing. AHK v2 strings are also UTF-16, so StrLen() and offsets always agree.
Line endings are normalized to bare LF on load because RichEdit stores CR-only internally. Feeding CRLF without stripping \r causes one-character offset drift per
line — the highlight gets progressively misaligned through the document.
The blinking caret is permanently suppressed by hooking WM_SETFOCUS and calling CreateCaret(hwnd, 0, 0, 0) + HideCaret each time RichEdit gains focus, plus after every programmatic SetSel call. A zero-width caret is invisible regardless of blink.
Drag-and-drop uses DragAcceptFiles + WM_DROPFILES. Only the first dropped file is used; non-.txt files are rejected with a message box.
Mouse-wheel direction on the WPM slider is flipped (wheel-up = increase WPM) via a WM_MOUSEWHEEL hook that only intercepts when the cursor is over the slider hwnd.
Center-scroll uses EM_LINEFROMCHAR, EM_GETFIRSTVISIBLELINE, and EM_LINESCROLL. Line height is measured from two currently-visible lines (not line 0/1) so the measurement stays valid after the view has scrolled away from the top.
June 2026 -- Based on suggestion by andymbody, added support for read aloud.
July 2026 -- Importand bug fix from rommmcek, see change log in forum.