Human Noise - Reverse Dead Internet

This project simulates realistic human writing traces and online activity schedules to confuse and pollute bot/AI detection datasets. By programmatically generating human-like noise, it creates authentic-looking activity across various platforms for privacy, security research, and data obfuscation.

How It Works

The tool is built on four core modules to mimic realistic human behavior:

1. Humanized Typing Engine

• Physical Proximity Typos: Simulates finger slips based on the physical layout of the keyboard. Wrong characters are chosen from adjacent keys.
• Layout-Specific Mapping: Adapts to different keyboards. For example, the German profile uses a QWERTZ layout (swapping Y and Z, adjusting neighboring keys), while other profiles default to QWERTY.
• Various Error Types: Supports substitutions (hitting the wrong key), transpositions (swapping adjacent letters), omissions (skipping letters), and double strikes (duplicating letters).
• Slang & Colloquialisms: Randomly replaces standard words with slang equivalents based on a configured rate (e.g., converting "to be honest" to "tbh", or Indonesian "yang" to "yg").
• Case Customization: Randomizes capitalization starting rules to mimic lazy mobile typists who omit capital letters at the beginning of sentences.

2. Markov Chain Text Generator

• Dynamic Generation: Generates randomized sentences using a local Markov Chain trained on target-language text corpora (stored in config.py), ensuring that generated posts/comments are coherent yet entirely unique and unpredictable.

3. Browser Simulation Telemetry

• Bezier Mouse Curves: Generates 2D coordinates (x, y) along a randomized cubic Bezier curve connecting start and end points to simulate human hand cursor navigation.
• Human Typing Delays: Generates time delays (with random Gaussian jitter) between keystrokes, with extra delays added for spaces and capitalization to simulate typing cadence.

4. Outbound Webhook Publisher

• Sends payloads to an active external URL via HTTP POST (using Python's native zero-dependency urllib).

Supported Profiles

• indonesian_netizen: Indonesian phrasing, local slang (wkwk, bgt, yg), QWERTY layout.
• english_netizen: English tech/Reddit phrasing, internet slang (lol, tbh, fr), QWERTY layout.
• german_netizen: German phrasing, German slang (zb, nich, u), QWERTZ layout.
• japanese_netizen: Japanese phrasing, Japanese internet slang (w, あり, おつ), QWERTY layout.

Project Structure

d:\humannoise/
├── humannoise/
│   ├── __init__.py
│   ├── config.py       # Profiles, keyboard layouts, slang dictionaries, and training corpora
│   ├── generator.py    # Text generation logic (typos, slang, casing, and Markov Chains)
│   ├── scheduler.py    # Circadian scheduling and timeline assembly
│   ├── exporter.py     # Platform-specific API payload formatters
│   ├── browser.py      # Human browser emulation telemetry (Bezier curves & typing delays)
│   ├── connectors.py   # Webhook publishing client
│   └── cli.py          # Command-Line Interface
├── tests/
│   ├── __init__.py
│   ├── test_generator.py
│   ├── test_scheduler.py
│   └── test_browser.py
├── pyproject.toml      # Modern Python package setup
└── README.md

CLI Usage and Output Examples

1. Custom JSON Profiles Loading

You can load additional netizen profiles or custom templates dynamically from a JSON file:

python -m humannoise.cli --custom-profile "/path/to/custom_profile.json" generate --profile french_netizen

2. Generating a Planned Timeline (with Markov Chain text generation)

Generate a static list of scheduled actions and target contents over a specific duration:

python -m humannoise.cli generate --days 1 --profile english_netizen --markov

Example Output:

[*] Generating human activity noise timeline for profile: 'english_netizen'
[*] Duration: 1 days
[*] Text Generation Mode: Markov Chain
------------------------------------------------------------

[Event #1]
Timestamp : 2026-07-15T09:24:30
Platform  : REDDIT
Endpoint  : https://oauth.reddit.com/api/submit
Content   : "ive been using thiis tool for years and had no idea it could do that shortcut. Mind blown."
------------------------------------------------------------

3. Running the Real-Time Daemon (with Webhook Execution)

Run a simulation loop that triggers events as time passes, logs simulated browser/mouse interactions, and sends HTTP requests to your webhook:

python -m humannoise.cli daemon --profile german_netizen --speed 3600 --markov --execute --webhook-url "https://your-webhook.site/path"

Example Output:

===================================================
   REVERSE DEAD INTERNET - HUMAN NOISE GENERATOR
===================================================
A tool to generate human-like online activity traces.
    
[*] Starting Human Noise Daemon using profile: 'german_netizen'
[*] Simulation speed multiplier: 3600.0x (1 second real time = 3600.0 seconds simulated time)
[*] Text Generation Mode: Markov Chain
[*] Execution Mode: ACTIVE (Post to Webhook)
[*] Target Webhook URL: https://your-webhook.site/path
------------------------------------------------------------
[*] Scheduled 4 events for the next 24 hours.
  - [07:48:44] on github
------------------------------------------------------------

>>> [EVENT TRIGGERED] 2026-07-15 07:48:44
Platform : GITHUB
Content  : Fehler beim Laden der Konfigurationsdatei: Dateipfad ungültig.
[*] Simulating user browser interactions...
    - Generated Bezier mouse curve starting at (277, 177) to (564, 443)
    - Coordinates preview (first 5 points): [(277, 177), (368, 173), (414, 185), (427, 211), (422, 246)]
    - typing delay preview (first 5 characters): [0.3758, 0.2152, 0.1375, 0.2994, 0.2372] seconds
[*] Executing webhook request...
[Success] Webhook response: HTTP 200: OK
------------------------------------------------------------

4. Running Unit Tests

Validate the stability of all modules:

python -m unittest discover -s tests

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0). See the LICENSE file for details.

Author

• fa33az