Wisper-Low

### Fast Windows dictation with a tiny overlay, local speech recognition, and cleaned-up output [](https://github.com/YashasVM/Wisper-Low/releases/tag/v0.1) [](https://github.com/YashasVM/Wisper-Low) [](#privacy) [](https://github.com/YashasVM/Wisper-Low/issues) **Local-first** · **Global hotkeys** · **Floating pill UI** · **AI text cleanup** ---

[!IMPORTANT] Prototype Release v0.1 Wisperlow is usable today, but it is still early. Expect rough edges with microphone devices, hotkey conflicts, and model latency on slower machines.

Download

Download the Windows installer from the GitHub release page:

Download WisperlowSetup-0.1.0.exe

After downloading:

1. Run WisperlowSetup-0.1.0.exe.
2. Follow the installer.
3. Launch Wisperlow from the Start Menu.
4. Press Ctrl+Alt+P or Ctrl+Alt+Space.
5. Speak, press the hotkey again, and Wisperlow pastes the cleaned text into the active field.

[!NOTE] Windows SmartScreen may warn because this prototype is not code-signed yet. Choose More info and Run anyway if you trust this build.

---

What Is Wisperlow?

Wisperlow is a Windows desktop dictation app focused on speed, privacy, and polished text. It listens when you trigger the hotkey, transcribes locally, rewrites the rough transcript into clearer text, and inserts it into the app you were already using.

Microphone -> Local Whisper STT -> Text cleanup -> Clipboard paste -> Active app

---

Features

Feature	Details
Global Dictation	Trigger dictation from anywhere on Windows with global hotkeys
Local Speech Model	Bundles a local Whisper model for offline-first transcription
Smart Cleanup	Improves punctuation, grammar, repeated words, and sentence flow before pasting
Floating Pill UI	Small black waveform pill while listening and processing
Active App Paste	Restores focus and inserts text into the currently active text field
Slash Modes	Supports /email, /professional, /casual, /slack, /short, and /raw
Cancel Flow	Cancel dictation without pasting using Ctrl+Alt+Backspace
Local Optional LLM	Can use a local Ollama model for stronger rewriting when available

---

Quick Start

1. Install

Download and run the installer:

WisperlowSetup-0.1.0.exe

2. Start Dictating

Place your cursor inside any text box, then press:

Ctrl + Alt + P

Speak naturally. Press the same hotkey again to stop recording.

3. Get Clean Text

Wisperlow processes the transcript, cleans it, and pastes the final result into the active text field.

[!TIP] For the best first test, open Notepad, click into the blank document, press Ctrl+Alt+P, speak one short paragraph, then press Ctrl+Alt+P again.

---

Hotkeys

Hotkey	Action
Ctrl+Alt+P	Start or stop dictation
Ctrl+Alt+Space	Alternate start or stop dictation
Ctrl+Alt+Backspace	Cancel dictation

If one hotkey is already used by another app, try the alternate hotkey.

---

Dictation Modes

Say a slash mode at the end of your dictation:

Mode	Use Case
/email	Polished email-style wording
/professional	Clearer workplace tone
/casual	Natural everyday tone
/slack	Short chat-friendly message
/short	Concise rewrite
/raw	Minimal cleanup

Example:

can you send over the updated prototype tomorrow morning slash email

Wisperlow rewrites it into a cleaner email-style sentence before pasting.

---

Architecture

Windows Hotkey
    |
    v
Overlay Bubble
    |
    v
Microphone Capture
    |
    v
Local faster-whisper Model
    |
    v
Cleanup + Optional Local LLM Rewrite
    |
    v
Clipboard Paste Into Active App

Runtime Components

Component	Current Prototype
App Shell	Python 3.11 background app
Overlay	Tkinter + Pillow-rendered high-DPI pill/blob
Speech-to-Text	faster-whisper local model
Text Cleanup	Deterministic rewrite pipeline with optional Ollama
Insertion	Clipboard paste with focus restore
Packaging	PyInstaller + Inno Setup

---

Privacy

Wisperlow is designed to run locally by default.

Data	Default Behavior
Microphone audio	Captured locally while dictation is active
Transcript text	Processed locally
Cloud services	Not used by default
Ollama rewrite	Optional and local when pointed at 127.0.0.1

---

Developer Setup

Use this only if you want to run from source instead of installing the Windows setup file.

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
python app.py

For a quick smoke test:

python app.py --self-test

---

Build Installer

The installer build uses PyInstaller and Inno Setup.

.\build.ps1

The output is written to:

release\WisperlowSetup-0.1.0.exe

---

Roadmap

• [x] Global hotkey dictation
• [x] Local Whisper transcription
• [x] Floating waveform pill
• [x] Processing pill
• [x] Windows installer
• [x] GitHub release asset
• [ ] Signed Windows installer
• [ ] Better first-run permissions screen
• [ ] Built-in model manager
• [ ] Usage dashboard
• [ ] Faster local rewrite model selection

---

Notes

• First launch may take longer while the local model warms up.
• Slower CPUs may take longer to process longer dictations.
• If speech dependencies are missing, Wisperlow should show an error instead of crashing.

---

**Made by [@YashasVM](https://github.com/YashasVM)** *Dictate anywhere. Keep it local. Paste clean text.*

Built With

• c#
• inno-setup
• powershell
• python
• rich-text-format