Voice2Text is a standalone Screph application for dictating and inserting text into compatible Windows applications. You can review the text before insertion. A window binding can also send a selected screenshot and the dictated request to a configured language model (LLM).
Video
Download
Extract the ZIP to a separate folder and run Voice2Text.exe. The application runs from the system tray and does not require an installer.
Start recording from the system-tray menu. A global keyboard shortcut for starting a recording is not currently available.
The speech is recognized by the selected service, after which the text can be checked and corrected. Ctrl+Enter ends the input, Esc cancels it.
Confirmed text is sent to the active compatible Windows application. Applications running at a different privilege level or using a nonstandard editor may reject the input.
A binding remembers the target window and insertion point. Before inserting text, Voice2Text checks that the correct window is active and that the point belongs to it.
A binding can include a screen region and an LLM profile. The model then receives a screenshot of that region and the dictated request; you review the transformed response before inserting it.
Menu and settings
Section
Opportunities
Tray menu
Start recording, show widget, bindings, settings, history LLM, language and exit.
Settings
Speech recognition, connections LLM, screenshots, snapshots, autoplay, clipboard and history.
Bindings
Several points for one window, image area, profile LLM and insertion method.
Recognition and data transmission
Option
Where is the data processed?
Vosk
Recognizes speech locally when the model is installed; Audio is not sent to the cloud service.
Yandex
Audio and request parameters are sent to the configured Yandex service.
Google Cloud Speech
Audio is transmitted Google with user credentials.
OpenAI-compatible
The audio is sent to the specified compatible service; the key is stored in the system secret store.
Processing via LLM
The text and selected photo are sent to the service from the active profile. Use only data that is allowed to be sent to this service.
Local data
Settings, local models, and optional file history are stored in the Voice2Text data folder beside the application.
Bindings are stored in the profile Windows in %APPDATA%\Voice2Text\bindings.json.
API keys are saved via Windows Credential Manager. The usual settings remain the service address, model and file paths.
Saving LLM requests and responses to a file is off by default. You can enable it and clear the saved history from the interface.
When persistent clipboard copying is enabled, recognized text enters the system clipboard and may remain in Windows clipboard history.
System requirements
Windows 10/11 x64.
A working microphone; window bindings and screenshots also require an interactive Windows desktop.
For local recognition, the Vosk model is needed. Cloud services require a network and user credentials.
