SuperScreecher9000

Voice control for your entire desktop. Local. Private. Free.

Press a key. Talk. It types. Or runs commands. That’s it.

The Problem

You want to talk to your computer. Your options:

Tool Cost Platform Setup System Control
Dragon NaturallySpeaking $300-700 Windows Installer + account Limited
Talon Free Cross Python + config + learning curve Yes (with scripts)
Voxtype Free Linux only Binary No
Nerd Dictation Free Linux only Python + VOSK No
SS9K Free Cross Binary Yes (TOML config)

Everyone else is either expensive, complex, platform-locked, or missing features.

What It Does

  1. Press F12 (or any key you configure)
  2. Talk
  3. Text appears at your cursor

Or say a command:

No cloud. No account. No internet required. Whisper runs locally on your machine.

Why It Exists

Accessibility tools should be:

SS9K is one file. One binary. Zero runtime dependencies.

Someone in an iron lung could use this. Someone with severe RSI could use this. Someone with ALS could use this.

They shouldn’t need $500 or a CS degree to talk to their computer.

Features

Dictation

Leader Word

SS9K uses a configurable leader word (default: command) to separate commands from dictation:

No reserved words. You can dictate anything. Change the leader in config:

leader = "voice"  # or "computer", "hey", whatever

Command Hotkey

Dedicated hotkey (F11 default) that auto-prefixes the leader word:

Voice Commands

Say “command” + any of these:

Category Commands
Navigation enter, tab, escape, backspace, space, up, down, left, right
  home, end, page up, page down
Editing select all, copy, paste, cut, undo, redo, save, find
  close tab, new tab
Media play, pause, next, previous, volume up, volume down, mute
Utility help (show commands), config (open config), repeat, repeat [N]
  languages (list all 99 supported languages)

Scratch That

Made a mistake? Say “command scratch that” to delete what you just typed:

Punctuation

Say “command punctuation” (or “command punk”) + symbol name:

Spell Mode

Say “command spell” + NATO phonetic or letters:

Input Output
command spell alpha bravo charlie abc
command spell capital alpha bravo Ab
command spell one two three 123
command spell alpha at bravo dot com a@b.com

Shift Mode (Selection)

Say “command shift” + direction:

Input Effect
command shift right Select 1 char right
command shift right times five Select 5 chars right
command shift word left Select word left
command shift end Select to end of line

Hold/Release (Gaming & Accessibility)

Input Effect
command hold w Hold W key (run forward)
command hold shift Hold Shift modifier
command release w Release W key
command release all Release all held keys

Hold mode rapidly presses keys together (rate configurable via key_repeat_ms).

Emoji

Say “command emoji” + name: smile, thumbs up, fire, heart, crab, poop, and 80+ more.

Case Modes

Say “command mode” + mode name to transform all dictation:

Mode Output
snake hello_world
camel helloWorld
pascal HelloWorld
kebab hello-world
screaming HELLO_WORLD
caps HELLO WORLD
lower hello world
alternating hElLo WoRlD
off hello world (normal)

Say “mode snake”, dictate naturally, say “mode off” when done. Great for coding.

Code Mode

Say “command mode code” to convert symbol names to tight symbols:

Input Output
function open paren x close paren function(x)
if x double equals y open brace if x == y {
array open bracket zero close bracket array[0]

Math Mode

Say “command mode math” to convert spoken math to symbols:

Input Output
one plus one 1 + 1
five times three 5 * 3
x greater than y x > y

Inserts (Text Snippets)

Define snippets in config, insert by voice:

[inserts]
email = "you@example.com"
sig = "Best regards,\nYour Name"
header = "// Created: {date}\n// Author: {shell:git config user.name}"
branch = "{shell:git branch --show-current}"

Say “command insert email” → inserts your email. Placeholders expand:

Wrappers

Define wrappers in config, wrap text by voice:

[wrappers]
quotes = '"'
parens = "(|)"
fire = "🔥"
div = "<div>|</div>"
Input Output
command wrap quotes hello world “hello world”
command wrap parens check this (check this)
command wrap fire awesome 🔥awesome🔥

Repetition

Mishearing Tolerance

Whisper mishears things. SS9K handles it:

Custom Commands

[commands]
"open terminal" = "kitty"
"open browser" = "firefox"
"screenshot" = "flameshot gui"
"lock screen" = "loginctl lock-session"

Any voice phrase → any shell command. Custom commands work without the “command” leader.

Alias System

[aliases]
"taping" = "typing"
"come and" = "command"

Whisper mishears something consistently? Fix it in config.

Flexible Hotkeys

VAD Mode (Voice Activity Detection)

For hands-free operation using Silero VAD:

activation_mode = "vad"    # switch from hotkey to VAD
vad_sensitivity = 0.9      # 0.0-1.0, higher = more sensitive
vad_silence_ms = 1000      # silence before processing
vad_min_speech_ms = 200    # ignore sounds shorter than this

Press hotkey to toggle listening. Speak naturally — VAD detects when you start and stop talking, then transcribes automatically.

Wake Word

Optional wake word filtering in VAD mode:

wake_word = "computer"     # only process speech starting with this

Utterances not starting with the wake word are silently ignored. Great for filtering background conversations.

Audio Feedback

audio_feedback = true      # beep on recording start, double beep when done

Logging

dictation_log = "~/.local/share/ss9k/dictation.log"  # log all transcriptions
error_log = "~/.local/share/ss9k/error.log"          # log errors

Hot-Reload Config

Install

Linux:

git clone https://github.com/sqrew/ss9k
cd ss9k
cargo build --release
./target/release/ss9k

Windows:

Requires Visual Studio Build Tools (C++ workload) and LLVM. Then build from x64 Native Tools Command Prompt:

git clone https://github.com/sqrew/ss9k
cd ss9k
cargo build --release

Pre-built binaries: GitHub Releases

GPU acceleration:

cargo build --release --features vulkan  # Linux/Windows (Intel/AMD)
cargo build --release --features cuda    # NVIDIA
cargo build --release --features metal   # macOS

Config

# ~/.config/ss9k/config.toml

model = "small"              # tiny, base, small, medium, large
language = "en"              # ISO 639-1 code
threads = 4                  # whisper inference threads
hotkey = "F12"               # any supported key
command_hotkey = "F11"       # auto-prefixes leader word
hotkey_mode = "hold"         # hold or toggle
toggle_timeout_secs = 0      # auto-stop in toggle mode (0 = disabled)
leader = "command"           # leader word for commands
key_repeat_ms = 50           # key repeat rate for hold mode
verbose = true               # set false once comfortable
audio_feedback = false       # beeps for recording start/stop

# VAD mode (optional)
activation_mode = "hotkey"   # "hotkey" or "vad"
vad_sensitivity = 0.9
vad_silence_ms = 1000
wake_word = ""               # e.g., "computer"

# Logging (optional)
dictation_log = ""
error_log = ""

[commands]
"open terminal" = "kitty"

[aliases]
"come and" = "command"

[inserts]
email = "you@example.com"

[wrappers]
quotes = '"'

Edit. Save. Changes apply instantly.

Comparison

Feature SS9K Talon Dragon Voxtype Nerd Dictation
Price Free Freemium $200-500+ Free Free
Fully Local Yes Yes Yes Yes Yes
Voice Commands Full Scripting Basic Punctuation No
Case Modes 12 modes Yes No No No
Code Mode Yes No No No No
Math Mode Yes No No No No
Insert Snippets Yes + {shell:} Python No No No
VAD Mode Silero Yes Yes No No
Learning Curve Low High Medium Low Low

TL;DR: Talon-level features without the scripting complexity.

Tech

No Electron. No Python. No frameworks. Just Rust.

Status

Built with Claude. Free forever.

Screech at your computer. It listens. 🦀

← Back to projects