Whisply

Whisply is a Windows dictation app that records from a global hotkey, transcribes speech locally with faster-whisper, and inserts the result into the currently focused app.

It is designed for low-friction everyday use:

• global hotkey
• local transcription
• tray-based controls
• first-run setup for CUDA runtime and recommended models
• installer build for non-technical users

Status

Whisply is currently focused on Windows desktop use.

Current product direction:

• faster-whisper on cpu and cuda
• OpenVINO backend code exists internally but is currently not exposed as a product path
• models are stored in an app-owned directory
• CUDA runtime can be downloaded on demand
• NVIDIA/CUDA is the primary GPU path; AMD systems currently fall back to CPU-oriented use

Features

• Global hotkey recording (win+ctrl by default)
• Hold-to-speak and press-to-speak modes
• Local speech-to-text with faster-whisper
• Model choices: small, medium, large-v3, large-v3-turbo
• Optional first-run setup for CUDA runtime and recommended models
• Configurable waveform overlay
• Tray controls for model, backend, microphone, settings, unload, and rescue copy
• Manual model unload from tray
• Rescue memory for the last dictation
• Bilingual app UI: German and English
• Optional file logging with a debug toggle in settings
• Packaged installer via Inno Setup

Downloads

For end users, use the GitHub Releases page.

Recommended download:

• Whisply-Installer.exe

Requirements

Runtime:

• Windows 10/11
• microphone input
• Python is not required for end users when using the packaged app

Development:

• Python 3.10+
• PowerShell
• Inno Setup 6 for installer builds

Installation

End users

1. Download Whisply-Installer.exe from Releases.
2. Run the installer.
3. Launch Whisply.
4. On first run, Whisply can suggest CUDA runtime and recommended models depending on detected hardware.

Developers

py -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
.venv\Scripts\python.exe -m pip install -r requirements-cuda.txt

Optional:

.venv\Scripts\python.exe -m pip install -r requirements-openvino.txt
.venv\Scripts\python.exe -m pip install -r requirements-optional.txt

Run from source

.venv\Scripts\python.exe main.py

Guide

For development, packaging, and troubleshooting details, see GUIDE.md.

Build

Installer only:

powershell -ExecutionPolicy Bypass -File .\release.ps1 -Target installer

Everything:

powershell -ExecutionPolicy Bypass -File .\release.ps1 -Target all -Clean

First-run behavior

On first launch, Whisply:

• checks available hardware
• chooses a sensible default backend/model profile
• can offer CUDA runtime setup when NVIDIA hardware is detected and CUDA is not ready
• can offer recommended model installation

Installed models are stored under:

• %LOCALAPPDATA%\Whisply\models

CUDA runtime downloaded by Whisply is stored under:

• %LOCALAPPDATA%\Whisply\cuda_runtime

Config is stored under:

• %APPDATA%\Whisply\config.yaml

Logs are stored under:

• %LOCALAPPDATA%\Whisply\logs

Privacy

Whisply is built for local processing.

• audio is processed locally
• transcriptions are not sent to a cloud service by default
• technical logs do not intentionally store full transcription content
• recordings are not automatically saved as audio files

Troubleshooting

Common checks:

• confirm the microphone is available in Windows
• confirm the focused app accepts normal paste input
• if CUDA is selected but unavailable, switch to auto or cpu
• if a model is missing, install it from the tray or first-run setup
• enable the debug log toggle in settings when you need deeper diagnostics

Screenshots

Current repository screenshots:

• tray menu
• settings window
• first-run setup dialog
• recording overlay

Contributing

Issues and pull requests are welcome.

If you contribute:

• keep Windows behavior stable
• avoid breaking the installer flow
• test both cpu and cuda paths when possible
• keep docs in sync with actual behavior

License

This project is released under the MIT License. See LICENSE.