Sotto

Local speech-to-text transcription for Linux/Wayland using Whisper models.

Sotto runs entirely offline — your voice never leaves your machine. It uses whisper.cpp for fast, local transcription.

Demo

https://github.com/user-attachments/assets/e8480c00-81dc-45d3-8369-d4e12b32ea1d

Features

• Fully local — no cloud services, no API keys, no internet required
• GPU accelerated — Vulkan support for NVIDIA, AMD, and Intel GPUs
• Voice activity detection — automatically filters silence
• Auto-paste — transcription typed directly at cursor via wtype
• Push-to-talk mode — hold a key to record, release to transcribe (requires input group)
• Spoken punctuation — say "period", "comma", "question mark" etc. to insert symbols
• Visual indicator — layer shell overlay shows recording time and status
• 12 Whisper models — from Tiny (78 MB) to Large-v3 (3.1 GB)

Installation

Arch Linux (AUR)

paru -S sotto-bin

AppImage

Download from Releases, make executable and run:

chmod +x Sotto-x86_64.AppImage
./Sotto-x86_64.AppImage

From source

sudo pacman -S gtk4 libadwaita gtk4-layer-shell pipewire wl-clipboard wtype vulkan-headers
cargo build --release
./target/release/sotto

Quick Start

1. Launch sotto to open the control panel
2. Download a model via "Manage Models"
3. Select your input device and language
4. Choose activation mode (Toggle or Push-to-talk)
5. Enable the daemon toggle
6. Configure your hotkey (see below)
7. Press the hotkey to record, speak, then release/press again to transcribe

Activation Modes

Sotto supports two activation modes, configurable in the control panel:

Toggle Mode (default)

Uses compositor keybindings to send a signal. Press once to start recording, press again to transcribe.

Hyprland (~/.config/hypr/hyprland.conf):

bind = $mainMod, V, exec, pkill -USR1 sotto

Niri (~/.config/niri/config.kdl):

binds {
    Mod+V { spawn "pkill" "-USR1" "sotto"; }
}

Sway (~/.config/sway/config):

bindsym $mod+v exec pkill -USR1 sotto

Push-to-Talk Mode

Hold a key to record, release to transcribe. No compositor configuration needed. Requires user in input group:

sudo usermod -aG input $USER

Log out and back in for changes to take effect. Available hotkeys: INSERT (default), SCROLLLOCK, PAUSE, F13-F24, RIGHTALT, or any custom evdev key name.

CLI Usage

sotto              # Open control panel
sotto daemon       # Run daemon directly
sotto enable       # Enable systemd user service
sotto disable      # Disable systemd user service

Dependencies

Runtime	Purpose
gtk4, libadwaita	Control panel
gtk4-layer-shell	Visual indicator overlay
pipewire	Audio capture
wtype	Auto-paste transcription
vulkan-icd-loader	GPU acceleration

Models

Models are downloaded via the control panel and stored in ~/.local/share/sotto/models/.

Model	Size	Notes
Tiny / Tiny (EN)	78 MB	Fastest, lower accuracy
Base / Base (EN)	148 MB	Good balance (default)
Small / Small (EN)	488 MB	Better accuracy
Medium / Medium (EN)	1.5 GB	High accuracy
Large v1/v2/v3	3.1 GB	Best accuracy, slower
Large v3 Turbo	1.6 GB	Fast + accurate

English-only models (EN) are smaller and optimized for English speech.

Spoken Punctuation

Say punctuation out loud and it will be converted to symbols:

Say	Insert
period, comma, colon, semicolon	. , : ;
question mark, exclamation mark	? !
open/close paren, bracket, brace	() [] {}
new line, new paragraph, tab	newlines, tabs
dash, hyphen, underscore	- _
hash, asterisk, slash, pipe	# * / |

License

MIT