HoldToType — Settings

Summary

—

hold it and speak — the text lands where the cursor is

—

Quick settings

Models

Keep a history of dictationstext only, on this computer; audio is never kept

How long to keep

How many entries to keep

Never record from these programs

What you dictate into these programs never reaches the history. The pasting itself works as usual.

Entries

What was dictated: copy it, paste it into any window, or delete it.

Keyboard shortcut

Modehold the keys, or press once to start and once to stop

Min recording, msignores accidental key presses

Max recording, s

Press Enter after paste (auto-submit)sends the message right away

Restore clipboard after pasteimages and files come back as they were

Insert character-by-character instead of the clipboardhelps where a field refuses a clipboard paste

Delay before insertingwhen the program is not ready for the text yet

On-screen overlay

during dictation the screen shows that recording is on

Show the plate

Your own spotthe plate can be dragged anywhere

Where to show the plateclick the screen — the plate lands there

At the cursorat the cursor — next to where you type; if the app hides it, next to the mouse pointer

Show the recognised texton the plate after insertion, instead of the character count

Microphone

Input level

Check the microphonethree seconds of recording, then a verdict: level, clipping, whether there is speech

Recording sound cues

Cue sound

Source speech languageyou speak it; it decides the recognition model

Change the text output language

Default text output languagethe text is translated into it; the plate dialog offers it first

Ask for the text output language

Timeout, sec

Which model serves which language

The Auto-detect model serves every language that has none of its own — those are the grey rows. Click a language to give it a model of its own. The same model works out which language is being spoken when no source language is set.

Your own modelAdd a local speech recognition model

1

Open the models folderDestination folder: models\

2

Copy the modelPick one of the supported layouts

Whisper (GGML)One file

ggml-*.bin

Sherpa-ONNXA folder with the model files

encoder.onnxdecoder.onnxtokens.txt

3

Restart the app✓ The model appears for the languages it supports

Unload from memoryfrees the memory; the next dictation loads the model again

Check installed modelscompares model files with reference hashes

Punctuation and capitalswhere punctuation and capitals come from

Dictionary

Comma-separated terms, names and abbreviations — a hint for the ear, not commands. It works for Whisper; Russian speech through GigaAM ignores it. The default set follows the recognition language until you write your own.

Replacements after recognition

What was misheard becomes what you meant — right after recognition, before the prompts. Applied from top to bottom.

Voice commands

What you say turns into a line break, a symbol or a cancel instead of landing in the text. Matched as whole words, applied top to bottom, after the replacements.

Turn on post-processing

Edits the recognized text by your prompts: removes filler words, fixes punctuation, changes style. Off — the text is inserted as recognized.

Model

LocalHugging Face ↗

External server

Empty by default — all post-processing runs locally. Enter an address and the prompts run on an external server: OpenAI, Groq, your own vLLM — anything with a compatible API.

Prompts

Checked profiles apply one after another, top to bottom, on regular dictation. Nothing checked — text is inserted as is.

UI language

Designfont, shape, effects and motion

Colourthe colour of the window, the plate and the tray icon

Start with WindowsAn entry in the current user's startup list

Check on startupthe only network request besides model downloads

Updates

Logeverything the app writes about itself

Re-read config.jsonif you edited the file by hand

Reset the settingseverything back to factory, except models and history

Recognition server

Whisper models are run by a separate program. It can work on this computer or on another one — pick which one to use.

On this computer

On another computer

The same whisper-server, running somewhere else: a home server, a machine with a graphics card, the computer next door.

HoldToType

Voice → text at the cursor position.

External dependencies

The libraries and programs this app is built from, and their licences. The address under each one opens the project page.

How it works

Hold the shortcut, say a phrase, let go — the text lands at the caret.

Overlay

Every answer carries a number: 1…9 pick one, Enter takes the highlighted one.

Tray and files

config.json — all settings; edits made by hand apply through Re-read in the System section.

Install and portability

The installer downloads nothing by default.

Get in touch

A bug, an idea, a question about a setting — write an email if it is personal, or open an issue if it is a bug.

Email

Repository

Bugs and ideas issues