From eb2c6f91de0836f72ab385162bb3ea6e6906421a Mon Sep 17 00:00:00 2001 From: Win Dictation Dev Date: Thu, 11 Jun 2026 21:42:41 +1200 Subject: [PATCH] Gitea Pages: user manual and architecture docs --- architecture.html | 520 ++++++++++++++++++++++++++++++++++++++++++ index.html | 567 ++++++++++++++++++++++++++++++++++++++++++++++ screenshot.png | Bin 0 -> 19589 bytes 3 files changed, 1087 insertions(+) create mode 100644 architecture.html create mode 100644 index.html create mode 100644 screenshot.png diff --git a/architecture.html b/architecture.html new file mode 100644 index 0000000..46249d8 --- /dev/null +++ b/architecture.html @@ -0,0 +1,520 @@ + + + + + +Win Dictation — Architecture & Engineering Review + + + + + + + +
+ +
+

Architecture · Code Analysis · Engineering Verdict

+

Win Dictation: under the hood

+

A guided walk through a native Win32 C++ speech-to-text app — how it's built, why it's built that way, and an honest assessment of the code for anyone picking it up for the first time.

+
+ ~3,400 lines C++ (app) + Win32 + GDI+ + SDL2 + whisper.cpp + Single .exe, no runtime deps + Target 2-core i5, CPU-only +
+
+ + + + +
+
01

The verdict, up front

+
+

Bottom line

+

A genuinely strong, characterful single-purpose tool that punches well above hobby grade.

+

The hard engineering — the architecture choice, the self-calibrating progress estimator, the seam-free single-surface renderer — is thoughtful and well-executed. The app does exactly one thing and does it well on hardware most tools would choke on.

+

What holds it back is organizational debt, not algorithmic weakness: a 1,500-line main.cpp, a layer of vestigial child-window controls left over from an earlier design, and a pile of stale documentation that describes a GPU-streaming app this no longer is. None of it breaks the running product — but all of it raises the cost of the next person walking in.

+
+

The sections below back up every part of that judgement with specifics from the source.

+
+ + +
+
02

Stack at a glance

+

Deliberately lean. No UI framework, no managed runtime, no garbage collector — just the OS and three libraries.

+
+

Language

C++ (MSVC, Release /O2 /GL /LTCG, AVX2/FMA/F16C)

+

UI

Raw Win32 + GDI+ immediate-mode painted surface

+

Audio capture

SDL2 16 kHz mono, F32

+

Inference

whisper.cpp whisper_full, CPU backend

+

Networking

WinHTTP model downloads, system proxy aware

+

Persistence

Plain INI + UTF-8 text files no database

+

Build

CMake + a PowerShell convenience script

+

Footprint

One .exe + a few DLLs + model tiny RAM, no install

+
+

The whole product is native code with no framework abstraction between it and the Win32 API. That's the source of both its biggest strength (a tiny, fast, dependency-light binary) and its biggest cost (everything is hand-rolled, including the widgets).

+
+ + +
+
03

Architecture & data flow

+

A linear pipeline with a single inference step. Audio in, text out, no streaming loop.

+
+

trigger

Hotkey

Global RegisterHotKey. Captures the previously focused window.

+

capture

SDL2 mic

16 kHz mono into an in-memory buffer. Near-zero CPU.

+

buffer

PCM in RAM

Accumulated under a mutex; RMS energy tracked for the meter.

+

inference

whisper_full

One pass on a worker thread when you stop. Trim → transcribe → clean.

+

deliver

Insert + paste

Text to caret, to clipboard, into the prior window.

+
+

Two side channels run alongside the main pipeline: a progress estimator that predicts and smooths the transcription countdown, and a timing model that learns this machine's speed and feeds back into the next prediction. Results return to the UI thread exclusively via PostMessage; shared flags are std::atomic.

+
+

Mental model

+

Think of it as a tape recorder with a transcription button, not a live captioner. The architecture has no per-frame transcription loop at all — which, on a 2-core CPU, is the whole point.

+
+
+ + +
+
04

The defining decision

+

The single most important thing to understand: this app was rebuilt from a live-streaming design into a push-to-talk batch design — and that was the right call.

+

An earlier version used the classic whisper.cpp streaming approach: a rolling 5–6 second window re-transcribed every ~0.4 seconds. That technique assumes spare cores. On the target machine — an Intel i5-7th-gen with two physical cores — the buffer backlogged, audio was re-transcribed, and the UI starved. The symptom looked like "the model is slow"; the real cause was an architecture that needed hardware the target didn't have.

+

The fix wasn't a faster model. It was removing the streaming loop entirely:

+ + + + + + + +
AspectOld: streaming windowNew: push-to-talk batch
CPU while speakingPinned — constant re-inferenceNear idle — just buffering
Inference callsMany per secondExactly one, on stop
AccuracyLower — partial context windowsHigher — full clip, full context
UI responsivenessStarved under loadFree until the single pass
PredictabilityVariable lagPredictable few-second wait
+

This is textbook root-cause engineering: the team correctly diagnosed that the bottleneck was the shape of the work, not its size, and changed the shape. Everything else in the codebase — the batch worker, the progress estimator, the physical-core thread default — follows from this one decision.

+
+ + +
+
05

Threading model

+

Four threads, one rule: only the UI thread touches the UI. Everything else reports back by message.

+ + + + + + +
ThreadLifetimeJob
UI threadWhole appMessage loop, all painting, the 16 ms animation timer and 50 ms update timer.
Model preloadDetached, onceLoads the Whisper context off the UI thread at startup so the window appears instantly.
Transcribe workerPer clipRuns whisper_full; posts WM_APP_PROGRESS during and WM_APP_RESULT when done.
DownloaderPer downloadWinHTTP fetch on its own thread; posts WM_APP_DLPROGRESS.
+

Cross-thread state is handled with discipline rather than locks where possible: std::atomic booleans (m_recording, m_busy, m_abort, g_modelLoaded, g_modelOk) gate state transitions, and the only shared buffer — the captured PCM — is protected by a dedicated mutex. The audio callback (driven by SDL's own thread) appends under that mutex; stop_and_transcribe swaps the buffer out under the same lock before handing it to the worker. That swap-not-copy handoff is a nice touch.

+ 
// transcriber.cpp — physical cores, not logical, by design
+int Transcriber::default_threads() {
+    unsigned hc = std::thread::hardware_concurrency();
+    if (hc <= 2) return (int)std::max(1u, hc);
+    return (int)(hc / 2);   // 4 logical → 2 worker threads
+}
+

Defaulting to physical cores rather than hardware_concurrency() is the correct choice for compute-bound SIMD inference — hyperthreads contend for the same execution units and would only add scheduling overhead.

+
+ + +
+
06

Source map, file by file

+

The app is small and mostly header-only outside the two big translation units. Here's where everything lives.

+ + + + + + + + + + + +
FileRoleNotes
main.cppWindow, painting, interaction, settings view, clipboard, paste, model selection, popups~1,500 lines. The monolith — see §13.
transcriber.{h,cpp}SDL capture, Whisper preload/inference, progress & abort callbacksClean, well-scoped class. The model layer.
timing.hPer-model least-squares timing model + live progress estimator + INI persistenceThe standout module. See §08.
history.hSession text files, UTF-8 r/w with BOM, index, pruning to 100Self-contained, header-only.
downloader.hWinHTTP model downloader on a background thread.part + atomic rename, cancel, proxy-aware.
stats.hLifetime usage totals + derived figures (wpm, real-time factor, time saved)INI-backed, header-only.
settings.hApp settings read/write via GetPrivateProfile*Simple and transparent.
text_util.h · logging.hTranscript concatenation; timestamped file logTiny helpers.
tests/test_core.cppUnit checks: append logic, bad-model handling, real-WAV transcription, progress monotonicityModest but meaningful. Built as test-core.
+

The decision to make most subsystems header-only and independent (timing.h, stats.h, history.h, downloader.h, settings.h) is a good one for a project this size: each is cohesive, individually readable, and free of cross-dependencies. The contrast with main.cpp — which absorbs everything else — is stark.

+
+ + +
+
07

Deep dive: the single-surface UI engine

+

There are no buttons. Everything you see is painted onto one double-buffered surface — and that's a deliberate fix, not a shortcut.

+

The previous UI composited around nine separate themed child windows (buttons, statics, an edit). That produced thin hairline seams around every control — the hard-edged holes WS_CLIPCHILDREN punches per child, plus the edit's themed border. Rather than chase pixel borders, the rebuild eliminated the cause: collapse the controls into one painted region.

+

The model is a small immediate-mode system:

+
    +
  • A flat Widget g_w[] array — each entry is a kind, a rectangle, and hover/pressed/anim state. No HWNDs.
    • +
  • LayoutWidgets() positions them; PaintSurface() draws each into an off-screen DC with GDI+, then blits once (no flicker).
    • +
  • HitTest() maps a click point to a widget; OnClick() dispatches the action.
    • +
  • An animation clock eases each widget's anim toward a target (hover 0.6, active 1.0) on a 16 ms timer that stops itself when nothing is moving — no idle CPU burn.
    • +
+

Two "views" — Main and Settings — render onto the same surface, toggled by SwitchView(). Dropdowns (mic, history) are the one exception: they're real top-level WS_POPUP windows, because a surface-painted dropdown would render behind the transcript edit (a child HWND always paints above its parent's surface). That's a correct, well-reasoned exception.

+
+

A sign of maturity

+

The popup code carries a comment never to open a MessageBox from inside it — because WA_INACTIVE self-destroys the popup mid-handler, causing a use-after-free. Recognising that class of Win32 lifetime bug, and the GDI+ "GetHDC locks the Graphics object" trap documented elsewhere, shows real depth.

+
+

The one real child window that survives is the transcript EDIT — kept because a hand-rolled text editor with selection, scrolling, IME and undo is genuinely not worth rebuilding. Pragmatic.

+

The cost of this approach

+

Custom-painted controls are invisible to screen readers and UI Automation, and the app explicitly hides focus rectangles. The transcript box is accessible; the buttons are not. For a personal productivity tool this is a defensible trade, but it's the kind of thing worth stating out loud.

+
+ + +
+
08

Deep dive: the progress estimator

+

The crown jewel. Most apps fake a progress bar; this one runs a small statistical model that learns your machine.

+

Whisper only reports coarse progress (per 30-second chunk), so a naive bar jumps — 0, 34, 72, 100 — and a naive "time left" computed from stale percentages actually counts up. timing.h solves both. It has two parts.

+

1 — A learned timing model

+

Processing time is modelled as a linear function of audio length, proc = a + b·audio, fitted by decayed online least-squares. Each completed transcription feeds back a real sample; older samples decay (factor 0.97) so the model tracks the current machine state. Defaults are seeded per model family (tiny/base/small) so even the very first clip has a sane estimate, and the accumulators persist per-model in the INI.

+ 
void add_sample(double audio_sec, double proc_sec) {
+    const double decay = 0.97;
+    n*=decay; sx*=decay; sy*=decay; sxx*=decay; sxy*=decay;
+    n+=1; sx+=audio_sec; sy+=proc_sec;
+    sxx+=audio_sec*audio_sec; sxy+=audio_sec*proc_sec;
+    recompute();   // closed-form slope/intercept
+}
+

2 — A live estimator that only counts down

+

On stop, begin(predict) seeds a predicted total. As Whisper reports chunk progress, on_whisper() folds it in as a measurement via an EMA (α = 0.5) — nudging the estimate without the jumpy jumps. Meanwhile tick() advances a displayed "remaining" value that is strictly monotonic downward, with a clamped catch-up rate so it can speed up but never lurch backward, easing to 95% and snapping to 100% only on the real result.

+ 
void tick(double dt, float& out_frac, float& out_remaining) {
+    t += dt; disp_rem -= dt;
+    double raw_rem = std::max(0.0, T_hat - t);
+    double err = raw_rem - disp_rem;
+    if (err < 0) disp_rem += std::max(err, -maxCatchUp*dt);  // catch up, never jump back
+    double frac = t/(t+disp_rem);
+    if (frac > 0.95) frac = 0.95;   // park at 95% until done
+    out_frac = (float)frac; out_remaining = (float)disp_rem;
+}
+

This is more thought than most commercial apps put into a progress bar, and the test suite even asserts the progress is non-decreasing. It's the clearest signal in the codebase that someone cared about the feel of the product, not just its function.

+
+ + +
+
09

Deep dive: the transcriber

+

The cleanest class in the project — a tidy boundary between the OS/model and the rest of the app.

+

Transcriber owns the Whisper context and the SDL device, and exposes a small, sensible surface: preload, reload, start_recording, stop_and_transcribe, cancel, plus state queries and two callbacks (result, progress). Inference parameters are configured sensibly for dictation — greedy sampling, no timestamps, no prior context, blank/non-speech suppression, temperature 0 — and an abort_callback lets a long transcription be cancelled mid-flight.

+

Two small details worth calling out:

+
    +
  • Silence trimming. Before inference, leading/trailing silence is trimmed from the clip — cheaper and more accurate than transcribing dead air. (Note: this trims the buffer; it does not auto-stop recording — see the doc-drift note in §13.)
    • +
  • Output cleanup. clean_text() strips Whisper's [BLANK_AUDIO] / [NOISE] artifacts and trims whitespace, so the user never sees model noise.
    • +
+

The class is also defensively coded: a missing model file makes preload return false cleanly (the test suite verifies this), start_recording bails if a device won't open, and clips under ~0.3 s short-circuit to an empty result rather than invoking the model.

+
+ + +
+
10

History, downloads & persistence

+

No database, no registry sprawl — everything is a file next to the executable. Transparent and portable.

+

Session history

+

Each session is one UTF-8 text file in history\, named by timestamp. The clever bit is live archiving: the first clip of a session creates the file; subsequent clips rewrite the same file with the full text. So a session is always one tidy, crash-safe file — not a scatter of fragments — and it appears in the History popup immediately. A g_sessionPath global plus a FinalizeSession() helper handle the edge cases (manual edits, typed-only sessions, loading an old entry without resurrecting it). The list is capped at 100 with automatic pruning.

+
+

A real bug was fixed here

+

An earlier version stamped every fresh transcription as "loaded from history," so the duplicate-guard silently skipped archiving — sessions never reached disk while the UI claimed "Saved." The fix (live archiving + a corrected guard) is documented and shows the team chasing subtle state bugs to ground.

+
+

Model downloads

+

The downloader is more robust than it needed to be, in a good way: it streams to a .part file then does an atomic rename on success (no half-files), honors the system proxy, follows the Hugging Face → CDN redirects, supports cancellation, allows only one download at a time, and sweeps up stray .part files at startup.

+

Settings, timing & stats

+

All three live in a single win-dictation.ini under different sections — app settings, per-model timing accumulators, and lifetime stats. Using the OS's own GetPrivateProfile* API means zero parsing code and a file a user can read and edit by hand. For an app of this scope, that's exactly the right level of machinery.

+
+ + +
+
11

Anatomy of one dictation

+

Following a single clip end-to-end ties the whole system together.

+
    +
  1. WM_HOTKEY fires → the app records g_prevForeground and the current selection (EM_GETSEL) so it knows where to paste and where to insert.
    2. +
  2. g_tx.start_recording() opens the SDL device; the audio callback appends PCM under the capture mutex and updates the RMS energy meter.
    3. +
  3. The 50 ms UI timer animates the level meter and ticks the on-screen recording clock.
    4. +
  4. Second WM_HOTKEY → g_est.begin(g_timing.predict(len)) seeds the progress estimate; g_tx.stop_and_transcribe() swaps the buffer to a worker thread.
    5. +
  5. The worker runs run_inference() → whisper_full. Whisper's progress callback posts WM_APP_PROGRESS; the estimator's on_whisper() EMA-folds it in.
    6. +
  6. On completion the worker posts WM_APP_RESULT.
    7. +
  7. The UI thread then, in order: snaps progress to 100%, records a real timing sample (add_sample + SaveTiming), updates lifetime stats, inserts the text at the saved caret with smart spacing via EM_REPLACESEL (undoable), archives the session, copies to the clipboard, and pastes into g_prevForeground.
    8. +
+

Every piece of the architecture shows up in that one trip: the atomics, the message hand-back, the estimator, the learned timing feedback loop, the editable transcript, the live history. It's a coherent design.

+
+ + +
+
12

What's done well

+
+

The architecture fits the hardware

Push-to-talk batch over streaming is the correct response to a 2-core CPU, reached by genuine root-cause analysis rather than knob-twiddling.

+

The progress estimator is exceptional

Decayed online least-squares + EMA fusion + a strictly monotonic countdown is far beyond what the task demanded — and it shows in the feel.

+

Seam-free UI by elimination, not patching

Collapsing nine child windows into one painted, double-buffered, DPI-aware, self-throttling surface removed the problem at its source.

+

Clean module boundaries (outside main)

Header-only, dependency-free subsystems (timing, history, downloader, stats, settings) are each individually readable and testable.

+

Robustness in the right places

Atomic rename downloads, crash-safe live history, graceful missing-model handling, cancellable inference, single-instance mutex, model preload off the UI thread.

+

Real product thoughtfulness

Smart insertion spacing, undoable edits, auto-paste into the prior window, auto-hide, learned timing, friendly stats. These are details a careful builder adds.

+
+
+ + +
+
13

What holds it back

+

All fixable, and none of it affects the running app. But it's exactly what a newcomer trips over.

+
+

main.cpp is a 1,500-line god object

UI, layout, painting, the entire settings screen, clipboard, paste mechanics, model selection, the popup window class, and stats formatting all live in one translation unit with dozens of globals. It works, but it's the hardest part of the codebase to onboard into. Splitting the settings view, the popup, and the painting helpers into their own files would pay for itself quickly.

+

Vestigial child windows & duplicate code paths

Startup still creates ~9 owner-draw child controls (record, pin, copy, paste, clear, two selects, two statics) and then immediately hides all but the transcript edit. Their dead WM_COMMAND handlers duplicate the painted-widget OnClick logic — e.g. the Copy action exists in two near-identical places. Leftovers from the rebuild that should be deleted.

main.cpp — CreateWindow(...) blocks then ShowWindow(..., SW_HIDE)

+

Documentation describes a different app

This is the most actively misleading issue. CUDA-SETUP.md, QUICK-REBUILD-GPU.md and FIXES-APPLIED.md describe a streaming, VAD, ring-buffer, 24-thread, RTX 3090 design that no longer exists. build.ps1 still hunts for CUDA and downloads base.en though the product is a CPU-only tiny.en app. TESTING.md references a test-audio.exe the CMake doesn't build (it builds test-core). A newcomer reading the docs would form a completely wrong mental model.

+

The README claims a feature that isn't there

Both README.md and CHANGES.md describe a "500 ms silence auto-end timer." The recording loop has no such logic — it only auto-stops at the 10-minute safety cap. (Silence is trimmed before inference, which is likely the source of the confusion.) Either implement it or remove the claim.

main.cpp WM_TIMER recording branch vs README "Audio Processing"

+

Heavy reliance on global mutable state

The UI is coordinated through dozens of file-scope globals (g_*), a mix of atomics and plain values. It's manageable at this size and the threading is disciplined, but it makes the code hard to reason about in isolation and easy to break with a careless edit.

+

Build declares C++11 but uses C++17

CMakeLists.txt sets CMAKE_CXX_STANDARD 11, yet the code uses std::size() (C++17). It compiles only because MSVC's default is newer. Set the standard to 17 explicitly so the build is honest and portable.

+

Minor: redundant color systems & no in-app hotkey editor

Three overlapping palettes coexist (CR_* COLORREF, T_* GDI+ Color, C_* aliases). And changing the hotkey requires hand-editing the INI — a natural gap given the polished Settings screen already exists.

+
+
+ + +
+
14

Recommendations

+

If the next session had a short to-do list, this would be it — ordered by payoff for effort.

+
    +

  1. Purge or archive the stale docs high

    Delete or clearly mark CUDA-SETUP.md, QUICK-REBUILD-GPU.md, FIXES-APPLIED.md, TESTING.md and DESIGN.md as describing the retired streaming design. This is the single biggest improvement to onboarding, and it's nearly free.

    2. +

  2. Reconcile the README with reality high

    Remove the "500 ms silence auto-end" claim (or implement it). Update the model table and build commands to match the CPU-only product.

    3. +

  3. Delete the vestigial child windows medium

    Remove the hidden owner-draw controls and their dead WM_COMMAND handlers so there's exactly one code path per action. De-duplicate Copy.

    4. +

  4. Break up main.cpp medium

    Lift the Settings view, the popup window, and the GDI+ drawing helpers into their own files. Even a mechanical split dramatically improves navigability.

    5. +

  5. Fix the build standard & align build.ps1 medium

    Set CMAKE_CXX_STANDARD 17. Strip the CUDA detection from the build script and default it to fetching tiny.en.

    6. +

  6. Add an in-app hotkey picker low

    The Settings surface already exists; surfacing the hotkey there closes an obvious UX gap and removes a troubleshooting step.

    7. +
+
+ + +
+
15

Scorecard & final word

+
+
Architecture & design9.2
+
Performance fit for target9.3
+
UX & polish8.7
+
Robustness & error handling7.5
+
Code organization5.5
+
Maintainability5.8
+
Testing4.8
+
Documentation accuracy3.8
+
7.1Strong, with cleanup debt.
An impressive core wrapped in organizational and documentation drift. The engineering earns a high mark; the housekeeping pulls the average down.
+
+
+

Final word

+

Win Dictation is a good codebase — at its core, an impressive one. The architectural judgement (batch over streaming), the standout progress estimator, and the seam-free renderer are the work of someone who diagnoses root causes and cares about how software feels. Those are the hard parts, and they're done right.

+

What separates it from "great" is entirely recoverable: a monolithic main file, dead code from a prior design, and documentation that actively describes a different application. A focused day of cleanup — most of it deletion — would lift this from "strong for its niche" to "exemplary small-app code." The good news for anyone inheriting it: the bones are excellent, and the to-do list is short.

+
+
+ Win Dictation — Architecture & Engineering Review
+ Native Win32 C++ · GDI+ · SDL2 · whisper.cpp · CPU-only · target Intel i5-7th-gen (2C/4T)
+ Assessment based on a full read of the current source tree. +
+
+ +
+ + + + + diff --git a/index.html b/index.html new file mode 100644 index 0000000..1f213cc --- /dev/null +++ b/index.html @@ -0,0 +1,567 @@ + + + + + +Win Dictation — User Manual + + + + + + + +
+ + +
+

User Manual · v3

+
+
+ +
+

Win Dictation

+
+

A push-to-talk speech-to-text utility for Windows. Press a hotkey, speak, and your words land in whatever app you were just using — fully offline, powered by Whisper.

+
+ Offline · runs on your machine + Hotkey Ctrl+Shift+Space + CPU-only · no GPU required + Whisper · tiny.en → small.en +
+
+ + + + + +
+
01

What Win Dictation is

+

A small, focused desktop tool that turns your voice into text anywhere on Windows — no browser, no cloud, no account.

+

Win Dictation sits quietly in your system tray. When you want to dictate, you press a global hotkey, speak a sentence or a paragraph, then press the hotkey again. A second or two later the transcribed text is copied to your clipboard and — by default — automatically pasted into whatever window you were using: your email, a chat box, a code editor, a document.

+

Everything happens on your computer. The audio never leaves the machine; transcription runs locally using whisper.cpp, a compact build of OpenAI's Whisper model. That means it works on a plane, behind a firewall, or anywhere with no internet at all.

+
+

Built for modest hardware

+

This build is tuned for an ordinary CPU-only laptop — the kind with two physical cores and no graphics card. It uses a fast, lightweight model by default and keeps your processor nearly idle while you speak, only working hard for a brief moment after you stop.

+
+
+ + +
+
02

Quick start in 60 seconds

+

Three steps. No setup, no sign-in.

+
    +
  1. +

    Launch win-dictation.exe

    +

    The window opens and a microphone icon appears in your system tray. Wait a moment for the status line to change from “Loading model…” to “Ready”.

    +
    2. +
  2. +

    Click into wherever you want the text, then press Ctrl+Shift+Space

    +

    Put your cursor in the email, chat box, or document first. Then hit the hotkey. The Record button turns red and a green level meter shows it’s hearing you.

    +
    3. +
  3. +

    Speak, then press the hotkey again

    +

    Talk naturally. When you’re done, press Ctrl+Shift+Space once more. A short progress bar runs, and your words appear — pasted straight into the app you were using.

    +
    4. +
+
+

That's the whole loop

+

Press to start, speak, press to finish. The text is on your clipboard and dropped into your previous window. You never have to click back into Win Dictation.

+
+
+ + +
+
03

The interface, explained

+

The whole app is a single window. Here is every control on it.

+ +
+
+
Dictation— ▢ ✕
+
+
+
Record1
+
2
+
3
+
+
Ready  •  2 threads4
+
Win-dictate is a desktop, whisper powered, speech to text application!5
+
+
Microphone (Realtek Audio)⌄6
+
History⌄7
+
+
CopyPasteClear8
+
+
+
+ +
+
1Record / StopThe big pill. Click to start; it turns red and reads “Stop” while recording. Same as the hotkey.
+
2PinKeeps the window always on top. Blue when active. On by default.
+
3Settings (cog)Opens the Settings screen for choosing and downloading models, and viewing your stats.
+
4Status lineShows “Ready” and the thread count when idle, a timer while recording, and a live countdown while transcribing.
+
5Transcript boxWhere text appears. You can edit it freely — click in and type, fix, or delete.
+
6Microphone selectorChoose which input device to record from. Opens a dropdown of all your mics.
+
7HistoryOpens a list of past dictation sessions you can reload — or delete individually.
+
8Copy · Paste · ClearCopy the transcript, paste it into your last window, or clear the box (saving it to history first).
+
+ +

The status line is your dashboard

+

That one line of dim text under the Record button tells you everything about the app's state:

+ + + + + + + + +
You seeIt means
Loading model…Starting up — the speech model is being read into memory. Wait a second.
Ready · 2 threadsIdle and ready to record. The number is how many CPU threads it will use.
Recording 0:14Listening. The timer counts how long you've been speaking. A green level meter pulses with your voice.
Transcribing 0:14 · 62% · 3s leftWorking on your audio. The bar fills and the countdown ticks down to zero.
Pasted / CopiedDone. Your text went to the clipboard (and into your previous window if auto-paste is on).
No speech detectedThe clip was silent or too short to transcribe. Nothing was added.
+
+ + +
+
04

How dictation works

+

Win Dictation is push-to-talk, not live streaming. You record a whole clip, then it transcribes the whole thing at once.

+

This is a deliberate design choice. Instead of trying to transcribe word-by-word as you speak (which pins a CPU at 100% and stutters on a modest laptop), Win Dictation simply records your audio cheaply while you talk, then does one fast transcription pass the moment you stop. The result is calmer, more accurate, and far lighter on your battery.

+ +

What happens when you record

+
    +
  1. You press the hotkey. The app remembers which window you were in, and where your text cursor was sitting inside the transcript box.
    2. +
  2. It records. Audio is captured at 16 kHz and held in memory. CPU use stays near zero. The level meter shows it's hearing you.
    3. +
  3. You press the hotkey again. Recording stops. Silence at the start and end of your clip is trimmed away automatically.
    4. +
  4. It transcribes. The full clip is run through Whisper once. The progress bar shows a smooth, self-calibrating estimate of how long it will take.
    5. +
  5. The text lands. It's inserted at your cursor (with smart spacing so words don't run together), copied to the clipboard, and pasted into your previous window.
    6. +
+ +
+

There is no “stop on silence”

+

Recording continues until you press the hotkey again (or click Stop). Pausing to think won't end the session — take your time. The only automatic stop is a safety cap at 10 minutes per clip.

+
+ +

The transcript box is editable

+

Unlike many dictation tools, the text area is fully editable. Click anywhere in it to fix a misheard word, delete a stray sentence, or type manually. When you dictate again, the new text is inserted at your cursor — so you can build up a document piece by piece, placing each new chunk exactly where you want it.

+
+ + +
+
05

Keyboard & hotkeys

+

Two global shortcuts work from anywhere in Windows, even when the window is hidden.

+
Ctrl+Shift+Space
Start / stop recording. The core hotkey. Press once to begin, once more to transcribe. If a transcription is already running, pressing it again cancels it.
+
Ctrl+Shift+H
Hide the window. Tucks Win Dictation away to the tray (and cancels any recording in progress). The hotkeys still work while hidden.
+
Esc (in a dropdown)
Closes an open Microphone or History popup without choosing anything.
+
Delete (in History)
Deletes the history entry you're hovering over.
+
+

If the hotkey doesn't work

+

You may see Hotkey in use — edit win-dictation.ini. That means another program already grabbed Ctrl+Shift+Space. You can change it by editing the hkMods and hkVk values in the win-dictation.ini file (see the FAQ).

+
+
+ + +
+
06

Auto-paste & the active window

+

The feature that makes Win Dictation feel invisible: it types into other apps for you.

+

When you trigger recording, the app notes which window had focus a moment before. After transcription, if Auto-paste is enabled (it is by default), it brings that window back to the front and pastes your text there automatically. You dictate, and the words appear in your email — you never touch Win Dictation's own window.

+

If you'd rather paste manually, turn auto-paste off in the tray menu. The text is always still copied to your clipboard, and the Paste button will send it to your last window on demand.

+
+

Pair it with Auto-hide

+

Turn on Auto-hide (tray menu) and the window disappears the instant it pastes. Combined with the global hotkey, dictation becomes a pure overlay: tap, speak, tap, and your words flow into whatever you're doing.

+
+
+ + +
+
07

History & sessions

+

Every dictation session is saved automatically, so you never lose a transcript.

+

A session is everything you dictate between clears. As soon as you finish your first clip, Win Dictation writes it to a timestamped text file. Each additional clip in that session updates the same file — so one session is one tidy file, kept up to date as you go.

+ +

Browsing and reloading

+

Click the History selector to open the list. Each entry shows its date, time, and a short preview of the text. Click one to load it back into the transcript box. The most recent sessions are at the top, and up to 100 sessions are kept (older ones are pruned automatically).

+ +

Deleting entries

+

Hover over any history row and a small ✕ appears on its right edge — click it to delete that session's file. You can also press Delete on the hovered row. The list stays open after each delete so you can tidy up several at once. The window even shrinks to fit as the list gets shorter.

+
+

Deletion is permanent

+

Removing a history entry deletes its text file from disk immediately — there is no confirmation prompt and no undo. The files themselves live in a history\ folder next to the program, if you ever want to back them up.

+
+
+ + +
+
08

Models & the Settings screen

+

A model is the AI that turns sound into words. Bigger models are more accurate but slower. Click the cog to manage them.

+

The Settings screen lists every model Win Dictation can use. Each row shows the model's name, file size, and a short hint. Installed models have a filled radio button you can select; ones you don't have yet show a Download button that fetches them directly from Hugging Face with a live progress percentage.

+ + + + + + + + + +
ModelSizeCharacter
tiny.en~75 MBThe default. Quick and light — ideal for this CPU.fastest
tiny.en-q8_0~42 MBSame speed, smaller file (compressed).fastest
base.en-q5_1~59 MBA noticeable accuracy bump for little cost.good balance
base.en~142 MBMore accurate; still reasonable on two cores.balance
small.en-q5_1~182 MBAccurate, but slow on this machine.slow here
small.en~466 MBThe most accurate offered — and the slowest.slowest
+ +

Choosing a model

+
    +
  1. Open Settings (the cog), and find the model you want.
    2. +
  2. If it isn't installed, click Download and wait for it to reach 100%. You can cancel mid-download, and only one downloads at a time.
    3. +
  3. Click the model's row to select it (the dot fills in).
    4. +
  4. Click Save. The app reloads with the new model and remembers your choice. Back or Cancel discards any change.
    5. +
+
+

A good rule of thumb

+

Stick with tiny.en or base.en-q5_1 for everyday use on a two-core laptop. Step up to base.en if you want better accuracy and don't mind waiting a beat longer. The small models are best reserved for short, important clips where accuracy matters most.

+
+ +

The progress bar learns your machine

+

You'll notice the transcription countdown is unusually accurate. That's because Win Dictation measures how fast your specific computer is with each model and remembers it. The more you use a model, the better its time estimates become — the bar counts steadily down rather than jumping around.

+
+ + +
+
09

System tray & options

+

Win Dictation lives in the tray. Right-click its icon for the quick options menu.

+
+

Auto-paste

Paste transcribed text into your previous window automatically. On by default.

+

Always on top

Keep the window above other apps. Mirrors the Pin button. On by default.

+

Auto-hide

Hide the window automatically right after it pastes. Off by default.

+

Exit

Fully quits the app. Closing the window only hides it to the tray — use this to stop it entirely.

+
+

Double-clicking the tray icon brings the window back. Closing the window with the ✕ doesn't quit — it just hides, so the hotkey keeps working in the background. Your window position, pinned state, and these toggles are all remembered between launches.

+
+ + +
+
10

Your statistics

+

Scroll down in Settings to see a running tally of your dictation habits.

+

Win Dictation quietly keeps lifetime totals and turns them into friendly figures:

+
    +
  • Total audio dictated and the number of clips.
    • +
  • Word count, plus your average speaking pace in words per minute.
    • +
  • Total processing time and your machine's real-time factor (e.g. “4× real-time” means it transcribes four seconds of audio every second).
    • +
  • Your longest single clip.
    • +
  • An estimate of the time you've saved versus typing at 40 wpm.
    • +
+

These numbers are stored locally and are just for your own curiosity — nothing is reported anywhere.

+
+ + +
+
11

Tips for best results

+
+

Click first, then dictate

Put your text cursor in the destination app before pressing the hotkey, so auto-paste knows where to send the words.

+

Speak in natural phrases

Whisper transcribes best with full sentences and natural rhythm. You don't need to over-enunciate or pause between words.

+

Pick the right mic

If accuracy is poor, check the Microphone selector — a headset or dedicated mic beats a distant laptop mic in a noisy room.

+

Match model to task

Quick chat replies? tiny.en. A careful paragraph of prose? Try base.en for fewer corrections.

+

Edit in place

Fix the odd misheard word right in the transcript box, then Copy — faster than re-recording the whole thing.

+

Let it warm up

The very first transcription after launch can be a touch slower as the model settles into memory. It's quick from then on.

+
+
+ + +
+
12

Troubleshooting

+ + + + + + + + + +
SymptomWhat to do
“Model not found”The selected .bin model file is missing. Open Settings and download a model (start with tiny.en), or place a .bin file in the models\ folder next to the program and restart.
“Hotkey in use”Another app owns Ctrl+Shift+Space. Change the hotkey in win-dictation.ini, or close the conflicting app. You can still record by clicking the Record button.
“Microphone error”The chosen input device couldn't be opened. Pick a different mic from the selector, make sure it isn't in use by another app, and check Windows mic permissions.
“No speech detected”The clip was silent, too quiet, or under ~0.3 seconds. Check the level meter moves when you talk, and confirm the right mic is selected.
Text pasted into the wrong placeAuto-paste targets whatever window was focused just before you pressed the hotkey. Click into your destination first. If in doubt, turn auto-paste off and use the Paste button deliberately.
Transcription feels slowYou're likely on a larger model. Switch to tiny.en or base.en-q5_1 in Settings. The small models are inherently slow on a two-core CPU.
Window vanishedIt hid to the tray. Double-click the tray icon, press Ctrl+Shift+H, or right-click the tray icon → Show Window.
+
+ + +
+
13

FAQ & where things live

+ +

Does my voice get sent anywhere?

+

No. All recording and transcription happen on your computer. The only time the app reaches the internet is when you click Download to fetch a model file.

+ +

Can it run completely offline?

+

Yes — once you have at least one model installed, no internet is needed ever again.

+ +

What languages does it support?

+

This build is tuned for English (the .en models). It's optimised for accuracy and speed in English on modest hardware.

+ +

Where are my files kept?

+

Everything sits next to win-dictation.exe:

+ + + + + + +
LocationWhat's there
models\Your downloaded .bin speech models.
history\One text file per dictation session.
win-dictation.iniYour settings, hotkey, window position, learned timing, and statistics.
win-dictation.logA simple timestamped activity log, handy if something misbehaves.
+ +

How do I change the hotkey?

+

Open win-dictation.ini in any text editor and edit the hkMods and hkVk values under [app] (they're standard Windows key codes), then restart the app. A built-in settings option for this is a natural future addition.

+ +

Why is it called “push-to-talk” if I'm not holding a button?

+

It's toggle-style push-to-talk: one press starts, another stops. You're not transcribing live as you speak — you capture a clip, then it's processed. This is what keeps it fast and light on a CPU-only machine.

+ +
+ Win Dictation · push-to-talk speech-to-text for Windows · powered by whisper.cpp (MIT)
+ Default hotkey Ctrl+Shift+Space · CPU-only build · runs entirely offline +
+
+ +
+ + + + + diff --git a/screenshot.png b/screenshot.png new file mode 100644 index 0000000000000000000000000000000000000000..1387eff21e953a865bfb532d75d5f24658c91fe4 GIT binary patch literal 19589 zcmeIaX*?9}_dks65|UJeP`1#95Tk5KXhX70vSyo*b(jf7@u|t4EF+TL*kzj`l{Ndm z&DgRHV;eJOy$7G~@4@}LAO2te`}g4cbY`yWT-Uj->wV67pL6EjJySzo?las>Oia8+ zxAh({F)`Dam{?dj*%_Y@Kiq9)Vv=Ao(!2R6&~9ytYgE)c9k)rZcL7dgb=nMAxMEt4 zOT6-k_E5s)Iquxi(PuxA?ET|vKodC-acS;IfSCTNQ+iIDE4DY(aB9fb))s7uSe)pYyQ(|AQsAq-_=s9ynqJ_QqT%OeK$LjJ#KI}@ z>XiVCq6QHdiZ_>%H$E?*!^F(Wf5}S63CT~nKww(=u>Yf{2~5qPN4^iOSos;>53q4U z2Ea^s!-JpRd4*9csLj84FxN5*D{5Km6-jD;?o;Ra)rka8>-4C0$1DBQV{-k|O8xa_ zRP-@C6zqg6vF-mtZ+u86Kw_&c+Ogw`Wc=2V*3)dpPKS%gGYlJT21U_6SBZCI{p#re z;}GWR=nJ`!9qAOGQTy`s2*Sc2bC8jXOEPw%-txnZX9X|){fxl$GW=s`$Lqt*-DfEa z2Lfwr@)byY(?BT#b)C?U0s_=2ZBk!d2udlMbo)J$rAzxtjYVhgIo-yj*59O;$cIqA~NGsghKh?MLnGt}CgV`4a{lL!^b0esW0hGH)|FIF{ zAAlv0QH?m`U_GqPpQVyIvU$Xk%?m9$H)dKKf0o+!_tn5euLoCl(YK|1`ng)$KUb|f z5=*^;p3NLr|C{%`+;D1<5cAPVQD0QHghvY6@TwEz&@>k zuud+wFxe$9wac{K@q4h*6}LFQbJg@=1tp+kP&fCsv3pD<-}TX!eyM+o3n_9w4xK^A z6-n9b!GjBKee)iIjIHZr+Jwn|5pzoM>5SbWcV~{QZiukmu7i-leGw>v?%qWIVvg?A z9Nd2=znrr0W@{b@k1f#-{R&>}g$Z|F_5q3Slyr=+auNtK$8A-$Wm3C1Lg*-pK}p9C4ngbd z>MYf)zPc=SW3N(I9JE-L_C%p+n4U=59-sPOrpNAlb#Vc-KdaT@XDMUWN1sK7_D!;k ztVJPu60uqL4W;{CFARq<4(*X;rhrU`J{MlL^O0p;pu0H_4QJ;U^tV;5@EQ-%(H|uv zUO*Z*$v55RnJ(b6|}*7QMphuKDoR4q&gW3Gb()IYdEMt%K}rl!cj});Dn|* z=^w$a5`j~!(BeAn34O*ZisRDTSAz6Ja1gqnr#-Y5Ool-@Xi~+{5KTszScx60)31xXyK`?dV+bzpqkQH!?D`-A zvFn05Cu2tYaKrYBBMuTS;hKleczWt(OlCvm-w6h&GqEfm3jb=EbbeMn0gepHi7IYsBXIZm_~J0ott~zRrml(v9_Ga)lPNG9abrb0xvYo&^&fbj9pK zr^vA95x@5!GZIhwF zfu)hiWGk;A!PHf@SA5M_qp=gT>Ycpg&5I9aP&H~&5$rpOM~dFq_bhDcx6qvQLsnZk;gn$64(f7>L% z9FETX{jK2ehy0YCgP$<79z8BWa9u&cwoFsINA^{rS|=y118%&|vW|B)ara~?|hB&oBJ+TTBT0EbM>=XEv) zc=ruI!!fwcu=aUn`SSbLp5e^^LJ!o%#2mrO35ht9S|E9_XT|`I|9b;6y~0u8L4p1; zm(ruBZdkA%RUgf~kh!lhohxB$spzCN>*?6(My>L7YEHYwkD7l3^Q<;t*D=@CiKV{t zP4C17&yqq8mfImXT5ez4K=>PurCTV;f}BF?pR*N+QLlz0%R* zT#9}53yQRuhB$a>ort1Cs9Ep0 zGKsTR>X?TE`4-@Nd*C86;ZSUWLa7VkWY7u>J-dRc!}`@%MRTB#KLBeLfcWa_k~;m& zr;N?K;$rDh`fh6XBru-KJm&*5ix4j)GzUt{+2#tZlh}V+GAuAO-Os8DOVg0!{gwAd z=T|roJuvd6X9J@6Ini_fY^GTn8?~;`Gk5|iJ(yz;$}oWK?SQMXJ(v&rLjpRP5VoCv z=wwjpIg<1%SP+X47bFx2rg+EF+4?YuDR}t3|_V+Envc z7_ZP?%$eJJEp=|5iEE6Tah2y{nx?fKlvd(g@@!~Dbw1Cmb0$ND*r!2fv2*BKqwSu4Mvm`;p&yoTFvgNbsfvH^Fkj8JxLp&wx zk57Wqhn{@KAd;?4+-wqMranPO(Q*1!`5WGZ_VX_AZ`IW527f}?R#$zMoI#NFAG>7& z@Jr;bKJO|l$WSbV^e4_-ETsSL>Ft|dDZ16elnwwML!VTG8{dcr0;lM!Ig(QmEvdKo zsPqZtIorL7X8I~!v3#`bU4*v$6*4QHPeBi9JN;BYuvt*)=7!AcId@J8j6$XiVTv5z zqL0Glu80d!;3B)?K4=Bujzwpagk%IgEsuE$yO=35h_aPGGVdMqdk<%>?AXRWq!ki+ z9`&rLbs>*-Q+Kb{nprcc?cqf3NdtDbNAO1xK){cHz5qKTn?&Wb^FHODIR)_B1)pzh z0_n;t{gRMov&x;OwhRo%in&6wx9K0n*K6oGHR_PI!X|`bZTl2VW-7n7l7j*JGJRJq)gmqUgvZUSgLP9pC#?0*ER^` z`tA9w`C;AlpL3rksUe6uUl^dsa3g@>Dobq)iX-BLl)K+;C%#PbV?`0TmE+B99`-Z? zjE-r$m5N^Rj>eYa7GgdRL5H6PX#w7u0hh(kS&Mi7S*bDQW^c1r045eIBdsb@X!-0= zb=c-Im~06YMFFTbynj$%Su{Ijuwqc{M_L9<*unt!S&&F7nW$})6>>h!c()RInOlg>A!ZuL=ar#)wKBDGGPi^%a+SFD zK=X&T+72%5r{$-eMGdYz`<}ZhL_RIj@l|1BT*wYTN^Av2PN7$}-N2({#nG4m-p%Br zeVUdf5!CmS#mp#m=G}J@tGM-A*1a!~#k-+!oDU3eho|Bw3+GRvz?pWDg*P%@Z^>0# z$$X+KBZ4%VG5eo_MFog{X4IOQ$%|j~*Q>iad!NgO8Rud4$^T25|4h#_p(Z;bKfWc@ z^86qD@AYZ**LG_d<SdoRmwIq_Uf22ur6~ZP%C(Y zBi7ibh9&DLUYHG_OWo9DMl&q%PnN2NR`dpJdm(CfXKw{>qYGR<;noyi|ivUn_jFhS4K?p{0wu?31P8kY-hon^>S zc!5@IeM{>T(_Y{@Mg5C>sO?uF>ND8&GrU#wo2ghg7ax>5KHZvO(r_C2mf=F2QZ8j! z$F1J<#WT2%?5`l2==^4vpnUE=|2R@o&SZW|c7K-V7< zZiUGeoA!_3f4{L0+V`Bs*xXdh5X(=aIzO)C%Y$C*E@*WJ>@3p3F;AScpL{#Ek`cU7 zQ{~sbSQ2cyV@>IxE@4J`2z(t>4Rrm=$lfdxj+(r(5xPIsX||o!U_4@b8Mr*>kfg)F z|Ggd9&XV}*OvJv|a}Unac@nJ@BaW=r<}u8dm%kT_8L^};0yThjhI>>_Fa6b@9S6}7 z3+wlBwR-R)X`)I#1VmRC21qar9N)eZqeJB!#Ewx(=PuPr3-LgTpL8pPy!R54hcP5P zy8Y@Ljk{U3*G=IJW&9?p9tdaJ*f(GNdBv|F_j|(8b{B*5;ILX2AwGynrz5~u^OTAq zQy7Uqkh)jzxCt*gzq6XH?)D_QfE88ph1Nd(P#Zt+6NE3!KQ6M*o}7841DslI8fos~ zU{<_Wg&7eUKx}DkeJvd1YywjdA-|xcDhU70$*7J|5?w%_|HdNbnvL z<4725E|AgLu`nU++vTD#8*(_i5G>`}+K^(ElUqUFM)S!(%}Zr^ut;L-q@k&{*_E(~4(v z^JzGnb5mxyjdqr&Ve|qU@bvs7Kq(nO#R!-GwB;@AZgGmGrEb&BD(b>k3*z*LR$3wt z4pjd9y)T%aH<`98(XZ`=ess$dzLd`Em%#Fkp-%FfT*jxtEV(&t~Nec*V=^+=X@WSwCX$*cU@=Bnc@%cj#!=Uy%vh5PC?uNF)W zId_wHLE4z_S$u92*!J+iKjxhBE`unmRMQs8%ruiBzkG(NWXyCtjGGNQBZfkfl+HvF z(kM0elQhDC&_(Rzr68KcE0w!STVx6q-`^M3)7I$ISqH0+Tx@*&%x*K;Ccl38w94d^ zjFwP&sh8iV;BhJVsPwXS2*>`T%Iss67o)_9ba(e4`O=n>dFQ>z$d~g9kY5>^mmine zN514Zw%zI>sh}4nl>RB9=t~hMz~proqTy&vmf!b9>7PY_n1lHGTB43nI55a+My}Ry z#)eRJ~_!=X)zi>%u20*?Z5(SFSQ1Y=J_{$FrlzVo>grW*_ctOV+{LrjdDV z11IFZmvbMy^Uqw5WQc3`rk|qC0=gXu;(vka1m@W7G|Qst3epUlGPZwF; z9KCc1Fn$WXqRk*PX9r2^DgSHShlHXy%O3#>-rh*=U%&!0@M}HwEz4ox`jLTO@Qm~B z2PCLY<~jqv>Ohj*`xcUMPW$*3oK(bh2v~It{8~l6la)A>w#)#{WvlWFhbXyX#{f<8 zg{)TxP{;X|0h+lk?Vtb83mv>jB)?>^-iA>-YZD^{Cf3&VB;4+f+wcF#(hUm*6O2GO z?LtSb`hd>qMsSWxkJuOr3N^oDJ%j;)K*?c8IE>H`sv{jqEeAmn*s+R{0#^_Sfd^+$@JoJsGP}CZBq$4j+twx(3u>dn%vwT-f@LS>0mW(c3HO zR<)|}vK><}R$ zXLl{?j_X00ONsfq}x15HL{@#vB;epB{?9^_qH!MB$Y3YQ&YdlM(PRv`E zbtKMfcK>jZhNusiD-``%4#=*s7xd~anFzFm$`(LbwqO4;+NmntzS`qRZa&M0_y-O4YUR-4Q=qgi~-CI%xAK7gg*LoLp2LDgR z2LoA$G}5)X@&zmbj}@rF{j$||g`2GYxcL4`PM^ULH-tK;eS~8tUd8P@U|FULBd+CE z>4B~_)^hmNt&Da0KrHF1Cp-x1^d8Fa#30%W+=bgtkEF}N3JSf*1()kXHc{vGxkz^E z9IvF$UTY7KzmKaNw)!C647X4y&G^2*)9(|A!7Xo?vUjo!ajwc{bY8nrRSWL5)Z4s8 z1LJ;M*E%c@hSQl;%0Symp81WHt*dvs$M=*5EJ@o7HY!DcnGZkiA$?zC0!j+!Z@EyB zZGM1+0beGV{&%6L=qGIur>e=I5&7DkILsH?EG`eBRpJGh)acWQJq?;eC{6w@^>Ri71WZo6K$BkVaw5{J9H&-z?8P7pM^{2`LUxU;AKQj^tPGiL)kKIv@6J~a zX#J_+@Tr%%-B+AVG5YEkZsl4bG4~Aj!!!4}C2g7%->8-AJ>x%or}hE*bKR=xX{eDs zcn^ZPydvoO49;pO3;h^M)0e$c?~@-NvjC8yc^vQiXerm zjJHlhC?rRi(YOZGV`CYtPt^#x{MYXW_c%POL@G~gT>wLYh?l}Rs?5&QPWt}>xO9uA zsQWr&zGbtb?3y=bbQh)qrj2agGt`(_=HNsnnY`nf#p|y&$&MZSj6N+xnIwS-=rsD; zF%2@xY@|6|CFf*T`CwubHN!AL87WVN@V)sTU#-p- zc{dgXYm%b1RxBM?9>TM5L1`!I3FKZ3-)co{SG1Yy8uFw?tL!oTR&~DkOb3TJYWrH^ z&L@jJGI0K4Xs24h%EY4F13O<%y>+wm~2|?*p1#1ql zR|jjYuE}G+jNb1clUlRkJK<>Lal&%|T0_Ik>)9xW$&cvD;>mZ9-sKuaXb6Tya+!>@v(je|EP>_H9MC35ng`9SJgSM8v zeCa%~nBY+}mj#g8UxV(3VPStt!63RZ937e??!RERVQpD3roa}O2z|3YKTM@!sf`#^ z?XlgB=V# zy;`(cGAL)^rLb%qTu6JyuN{44SJu7HQxstYU+)t~MU*(ydTb8nZ8a^$`GYY3CoQjVqfeOHa;RY~Hn-P*iV@DOW^c z0;=9?=NWGK{ONg-Y(#J{tKdY}LQyX)aA zMmCF2T7Ku~F3vOqDbnE;zIy7`{=l0-9MI)4?|X|M${iegzVi4=M0c#I$TZ35Y;z=B zDIvD_KXS2}YMEar0Q4>g0Ka?{qU`pPRA3o$BbtA|*;KEPx0X%=y39;Xj#Ip{B?U(~ zAA6_HU-(9{+2p`mBK$YOpTD^R-o5yR;KRl?hLlXUbcmgA$o^;N4KE(7ZvHhi>-y`p z+{?<*aagnUfP87KZwt!1(+^<2h3&n&+0*wwfb3fnuF=+oDR#^_wLTc8r|NZ;R7TaE zby4W^zKf96r=EYdMtt87Ss@%Ep_3Rwa61gm|IPRKo%XMVAHiAo*Q>}T;T>qS$)(uy z9UlZL{~2lSUG37uI^LvfwJEQ3TWb7c_i0i%cs8N_LA4B5P-Iq*76g@iXD?CHI*Ds* zcy6Vg#cu)lDdB5|3)LR z!;{W<4(q-2-i&o`#ISnH4FLQn-tlICc~of6wvpfnMBM)uu1eWFD>MrW$iQ<2bv{pE zaG}>yURa=f&)Tm&@QD?=5_|Pd+qcG}&G0N&nq5i)5;=kcJ-6AMFqcKvq*K%(+BKL7kEoABfTRrefC7(WK5{@e4WyOYB)eZTF(U(s=H)n-PUpRRyQ=9EaMmHmGu*{G0$w^BF z_vkeaUU1tRkOisKBpL(_Jc%;Lz4inZSUuUr*NKIr?5rbw&hg)rSP8QN4*k)bug(&W zi7q8YD<7|-JU{XE^TeI(hUdlEY1L{&7Zu6DwtUS%graUO|62)ud{f9wF*HHfOA!*O z@nb~+d3$gx_x?vWFNOD8->N?Nj88wRsKl=NX|X~ZtB)@H8#b~q;zYK4R_6}`AH2NI z#)y08J*C6QP+o$;pD|U%748I|yF1z`J)2}`#$IQFJWonzP7I#Wi*xxiHym+DUff~C zpX{qbiQLapzx1#6fMoufwW>b8df$dIRi@W|aO!ym7jCZcNud!Ej%wBr{iV+o_DxWN z_|T)DB9h{>dXZG@M^0`o0y@<{`n{MW7s``L%Bjk9N<(%1-So`}t~w7&^l5N=M4v*W zy|MiKm)oT={|^0*^>S(%3;IH`@*F|GWB`z#N<9(u;uri`kAwMs9Q=|w98hPDiQ7fs z#t`+2K?LwRK38L78wJEcKra)VZiS;9>q|y{MkWnVODL}_m;SEK1g{loxnV+JweYAB z(9_&O=_2<1xTHz<6~^p~wPS=5d17H#bIKTU@%LhKGdF5>$010r;|g;#4zCpGf|pFZ zDotD?AUSByCOgN_F3KeDqa(`iHlKYwp&?ed&o3V5L+`7;5S}T;SoSEQmaD zmqj8kVELCa$Z@){q?%MCE@-w)D$)VhH_2uS>H62a~Tyv zIYO;9Ie|WM8E|;&8`Qo+!p&KIE*Wn%q@D&zo-a^lB?>T2^b2g8xN;`x0|_N zJLKcCj;uzvHwSV(zOY+F^SB2j4HKMJiUo((*jv{)z^cfKw31w=hKyBZxtg9LS|-Mf z7cO)VX8$~ni3XfZrvLaw=&opPob3=^APtn$>AYFE&BETs1NVJKo!0L{)lHJyR;QKv ztmGcw>e)-Ciqh{7yF-fw_zAaJdNy+hw;%H|mdS7S3IkhA3|y9lA@r>-crz0v58n)< zs1Wf$?XYU+f%n)UO%~71Pce5JJt)qS7fSwfP+xH^pQ#!BmDjlHkXqSbltnBHAF>vw zq#0EegQJgx4)_}8aJcpXW&P^aO-7aC&G#neheQKoz>)tB_^+1sZ{B~e1}g)+{`(&O ziyHrh%KzGj0|Wbi!;1F4Xol|e8M-7jlImXKmy^Gz96O_FF^oVb5Xm6|2j)A>K*;qg z27b-HsPb;QhDwFckXtzFK3H_YU_XDHm#-tmMh|l^;HRjpDSC+R56PTiILMKpTk?lj z`Os?>GCoZ7^lmVG^cJ;qLWikoi8`ic@u}#4+aFv58A+j;)Q{W;3XtHpWq2uGzv2#^ zgli29hhxL!Hrt`Y@m7gZj#Bf?h#`=%kTZ<54~Tf->!BCJsmDkWWm}gq3Yhk1;#Jn! zPzw=!aOjS(X7OCJ*#J`B<)?t*7{lN!wafpm?H<4DcMPKzMFTxR(GU+kb+~^$7G7Zr zb6BPg6 zi7#dwA|qOP2MI@=3NGT*TtH-_tqWoXN>wW%rWRY?J1m|2CdB!9QrJHSU8+mv5Don3 zFcxLcaGQ)JJQ&H-*V>D9XV#;kJ4=_#J;|oaYil`{hnMhag3jX&!)C_K z5WoEF^$vD9kzVK(tL>Z5G*`Wm+q9YTF5o(&CKR^K zdv;v$^z+22`T!ZFqd>ibs*6kkwgMr_*|TT+a|>;Ui>;GCwR6)@D9TD#L;(7M#Rg$@ z#EkvuDQW8(PcxfS2Zch6@K1{Skphv7$H{;R*d$NunhVZ%6p8qa7Fj4&^DZCgfpd)uWko8SC;R#h{d|31L{rly-m92cvL%N-WG%iTV?ZTJoW zEvz@34&UGnwsK4Ik)kom0#aFqQMq~bSQM|wIqPaKh58qjfPY`++cPHO(y6K?%B$1J zwNK-SZFP}vKsv1-Mb{boeH2PkIQ*Rx*UsvMWuqtv<}p^DQ-ptXjfn5y;1iSd^gLVc zJ5ut9yW`Fw4qP(H%&vG?r}1Q{XBuHvydqRPJCfu6{CVs--KaZYu#6=wUq$$^x#p6- z4sf}q+S#!sk~e=^1Mzrcf;T&g%E5OgYkCf4ySp~0Mf>*h@yfgXNbciZ9kY3nYfoQc z#~Y=pW=cNvPn0?2V$mo-3Oo8={=tatC5Es%VW^=oKBR}_ zEw9hbl~gIdvoUcVhp=1KbltdzH-+8@{H?Hj&xD;b43+OMDmrlhnF6dk@h?cw^tHAA z@!>2xFd}2XQfhudyIlAYp@1D_`_WXz>I+if! zE;B^kg~TF%-a0&NiQZ2vV;DNtDJ9VtYOZp41-&b&%VJLok zsNH#gnC6#c&>I3Gz<*bInXTX?h2#=!{p3k8eAV+lK(C3BAnq4v%M)+x$Kktah%LaY zU&u7m`m&R(lPp8y$qHhW;$qJg4$eBq`K2R{Kim3NlA}lkgdIY~^bn)ciFi?eZu4xCiKpW2^Utq~(DtR!m@fhW1L}Xi~ zly(;64Z?80^UFU!Vz6YgIa-xbakj;}WAqZ}cYJCr-?YRqUvQOnWWLHtIUl}bqYNJ6 zdUfWAEa(Vn0^7MDQqInLi?yI!d9j00E@v6B*PlPla@~H`@wCccrd6Fvr|k=8#F$y- zFZKf2P}&Th{&_@EAk`cn1FY{jLvsO^yoe|ZW({HOpH5Cs&A+`7@Nze{{oyww7t7>^ z&2fct;8essqdIR$@YD6@(N%Os@B-Ud1=wLIEL*}o86fzoNfv*qIAe`I8JFvzHQS23TXQQD$r`u5 zpGVl;J*K`_S>ZTltbC~p2n>zvKP8{(!Y=N&m1QiM0XDOl+Qo+@}Au<7X0cR&{>I?5_$Yv znM%>G{fkNXQT|h^#AI^eY$mY5Hp%vi#k6^-l*q8tGhsGpm)~-ETCIeO_;$UJ-HrC< ztgUA8s(z;(SYGN=b=EU_675Kvk!EEc;f{YIER*eOmm@Y&a7iD`%`MYEB^y&O@qw56 zoj#v0+-5#?@3#+2w=ajcm%_d<9=2YYOBEJ9arHy(-6}qzdta=RVzh5I+RdYKU$De& z-IQ%_4#$|)3Oqi`BF8B`mK+*E?OKu8xLH%(; zjq*gzd>WopY^4UYxPY?9pKYIF@1hn0q4Hw^L*74YQPxvF9_=)TDwi`d|17HtyX!Vt zGPH{op#Vwqtyc+JTQA>7&}FS&IPY?oT$)7iopk94ZxL=w(I@PrbHw(O8-D4I)i%j% zTz52+O>3FTKzqN25zuF4js)k8c)xg84ME`+S9YrBg@rO1>9(~ z7Rk^mTQhdlij&zmbC_HX+Auo|FjhWu&@J)InB%Q4pfu$3-7eIn=WX7X0;kl^DOl>e z=VqkTt9-Qe_voWA!wPT0-r4G-8O14mp>)>O00_F^;j-gpy21&1;RHqEZ3W&|=H^)8 z;`NCEg>4?NrPZww&l*p!N{L%b;6G(=RcB{pQ-xGZJ}PO{a@U5=2#Lx?wzI*h4MG%U z)J|3xoE#Ohy4(v`PI$PWi#JXtOa>W#)pf6mKv(oED35DhSPHJvM>;o4u?;`_LU#7L z_WQD=#W+y%emOr&8Jk<9@(7fXxBc*DhK9X_B1iWNX@qg`2I#1p9mSrrO+7!aBYqNl zzS;ISsaRuAR9h?Ja!0BEI!5i`)R!0NE2S*TA>i{ZWzkAI(SVmzzglXi1#j5ej;wan zEjjk&PM!69IU!9b7Qiu;nW{FLDrw0CvQOH*3;OQr4iXM+`gF`E-o(wXE&=@|~~! zp*|<^U$yFOAV0CyMt)SH6UB8#1g~_jGjvRiXr|@t=G<4VacX6xC)66|nmKubf^2^ zK9azKde*l>Wg^P*5Kl8G~8bX8>#r@r!0MDFudm!hsN!0wr9&}IO`LCG81f2b8M zQ047`R8S#Ri$>`Ta&8|b49O>zQe8R^WxQw*yfr`i_KIow#Di+WdV87pE1DlxuTfWrVJjJ z&5&H(<*!|{rfW8cgoer1j6`^2+)Fc!moT@V&wFg-9;8h!7A0#xT{4RzDgKk!Lhi^(s zrYV&iF22a~Y03*dz8joohO&NY;KhtU37@@jCqk9tt&Ou^!nT_(fAq&tLEx{_boTm8 z@}Jrne3sz}-Ri|UOVmcNityZV{P2y{_K4Z`i#bmyX$v@-n;mInxnXs&jRn1T8!C8gq%k8HG`-3?l5}$ z;^G6*7{po;E!d-X=fZPyg#Rx=slpoD>I9)YAYGax#<3Yl5B>7ev53}rmrJ0OjbR;G zq4TtXx6MG5^2c{w(``Ik9n}Xk|#&HieX4VikjrzLzpQm#>!VSfu@f8i&&X@A6E5;JbNSnqxB9FoW&6ZTtkb zy_~q(`s&yX*M{K6CD+wgEer?SP3i@sb$l?C8oKp2dc+Zz;flitBqfB_s;>RE5O~4E za3mSBtp`tP)RWrp8;ZM3Ekfyy0j~NCCaJuscinzvgnBIgS4>7+#+0W16>kJVe>Ayp z($(V_Ghf)X-33N_=Cn@CnP*ruv{ZC2*5RumFG1OoKqjomXqUYmW;o=IEWx*5%$|Gb zxYTRE^`jzYXh6k!D7&h$-QqVgZS=jeG|adLUZjeI`*gu+OCgvMaD9*5X9)8F~< zleXk$GNXQzeGKWRAj7<{%ZY@&($x`{VCOq6AHl(wc;Dez>28&IVN$0ZzM^sAwKn<6 z)X}gN?nlAEYIg@VdY5mm%Q|!?u^c^wyU?%uhNNwerX5+3@MGyMjua>JosmAfc6MmU z(Vg>J*(-?E@bI5h$L%DOt!6N;gHiD6gtJpAQ*GWriVKo`1Op@rEYw(rMw9)ViZS$c ztiqL=>5zbxX565Mr>9`3s+1y$|HpnrbfWh({Jt(}nBH$w)>=23tZ=kz<^9RL-$)NO z`j{|8dwaAYO8yMRphi0J;GaWuGEG^FF!U~~fvel_yU;w8>B{ZPhrzUtgbbr?I=~q! zI~!_24iB{s!$HR9xlqcjzW`kDM@%l@UshzE zMF{)>)jpLcb{Nm|$}uQ7$BO%|hlAo6HQViz#FPJvzI2SBCzv@|C{cL;{qbENL5Jnd zy_XpM=shGxKYH2(Jst!4}3?lToD)(WpIHM`rV*?m;|L-?O%LB&Dxvl0<>8_`k6^N+IWa|_-n&KM{aISc- zMR!Y4X`I+BUf9YNr`$^VX~P#I#Sg^j?alHjeSDVzyjvljAv*wr8YEr(0KA2?&(I-W zzB~bPT1T~kmL96CN?2^beVQ$=82{}#N0d_2UC*o!`Z7t4M|W!0;+!-a+xmgi65Hqx z4OdR;Nrb(kGhdtFQ~Ubjc*jkRn=PkEiFvJhq(@pW`7`FsdbtNs6e)G^WACMjLN1bZ zxXGpvBMuEemihdw&A=1B_oG;rKG?gR{NN=XRw&luE!w*{{n-OAUHxp|C)r!yW%=Dwx3X9v7tLzdzAPDOEUg(EjAQ< z?#&q$syrv{yNQS2)=ar^Td_y7<7e zflr#1wYQmVs3WzCP8(V(&yl?ZWz0h6qnP%~tUvp_w!Ac@8 z^>cXU{7Tc$i4-{4x<^;W&b%|9B;gStCn@J#FbZkrzH_ceTah5wv37NC?(sDG_c?|}?(Ld=nr z(;IWy&;C*US#k>KDjSx^(Kz{HnSZe{5Qoz+JJxWlXjx0K_g?P_)khM)FZT&h%sy~} zNzz3)gDH+%3Qs2)^iuc*zB68gx+vgIms{5tNMYb`Vp8zk7wwnSOUPUo@aJ7C{*v$f zjK~;i&&7jwW#g=Ws~24RMoe*sw}1QNV*GxVTUz5nQkSx~r%ENxV3K3j;`gfsE}GtZ ztXBTdRdt$cJem9IUZZ%7++@!myLad1uL0fU84J}S+&E5_mshiS;sZmy+%hK&_`^Fx zEz@A-!CeoZ%e%-6nnggPPbqv)V|NZ9pu6Rmu zc8y*Wg1RUg;(2><%Ou=b1g#^%N?+wV8dGCW|SlvlR zyWaQ>`)1%N3#aQ<-*|uhW>9}&-vlIxso&)CIS(;(hH(tv1&sfPlxBqvTstpx*tj6U zXe8uQd&YH$y`K#jsglp%zDgeA*lR}R&5XgtfBC=xMlp}cK*A%|Lo9p2VB{v>p8pr8 z&M+4F{BvCNp%8vX0wzkNjP+2yi4+6iO&&fFI27`O!R)<|O8VD&JZF%O-=5!*XE-Gc zp*r6gIUlc|#}DOSd(TMgTr)A!KNP~r$lSENe|%7x^S^MDANgqC@&*3YD^r+G7@e