NOISE was written in July of 2013 as a way to create "real-looking" text based upon a collection of reference texts, which can then be used in emails, tweets, web searches, IRC chats, or any other medium you can think of that makes it a bit too easy to profile an individual's communication habits. Currently, NOISE only has email and Twitter dispatchers for generated texts.
You give NOISE one or more texts to use as a reference. The more texts, and longer texts, the better. NOISE then uses this collection of reference texts, called a corpus, to generate new, "real-looking" text, using an algorithm called a "Markov Chain". If you provide an optional set of keywords you want used in the generated text, NOISE will then replace any proper nouns in the generated text with a random selection of keywords from your list. NOISE will then email these generated texts to recipients of your choice, on a periodic but somewhat random schedule.
NOISE requires python version 2.7, the NumPy library, and the Natural Language Toolkit. See their websites for installation instructions:
- Python: http://python.org/download/
- NumPy: http://www.scipy.org/install.html
- NLTK: http://nltk.org/install.html (Requires
maxent_treebank_pos_taggerfrom NLTK Data)
For Debian-based systems, dependencies can be installed as follows:
sudo apt-get update && sudo apt-get install python2.7 python-pip
sudo pip install numpy nltk
python -m nltk.downloader maxent_treebank_pos_tagger
NOISE currently comes with five different programs, which can all be run on their own:
- noise.py: This is the main NOISE program, which uses noise_generate, noise_dispatch, and noise_tweet. It reads in a configuration file as its only parameter. This is probably what you'll want to run.
- noise_generate.py: This takes a collection of reference texts (called a "corpus"), and generates "real-looking" text based upon the corpus using Markov chains.
- noise_dispatch.py: The dispatcher simply connects to a mail server and sends an email based on the given parameters.
- noise_tweet.py: This program will tweet something on Twitter...
- noise_keyword_parser.py: This program takes lists of keywords as input and creates a weighted keywords file, to be used in noise.py or noise_generate.py.
NOISE comes with a default configuration file noise.default.conf that you can modify to meet your needs, and also has explanations of all the relevant options. NOISE can then be run as so:
python noise.py -c noise.conf
NOISE can also be run over Tor, in case you don't want your identity revealed. You must have Tor running with a SOCKS proxy in order to do this. Torsocks is the recommended way to run NOISE over Tor:
torsocks python noise.py -c noise.conf
NOTE: Tor does not provide perfect anonymity. Please see their full list of warnings before using Tor: https://www.torproject.org/download/download-easy.html.en#warning. Obviously, don't use a personally identifiable email address with NOISE if you want to stay anonymous.
It's up to you to provide your own corpus and keywords file. NLTK.org ships with some corpora you can use for free (see http://nltk.org/data.html).
NOISE is GPLv3 licensed. See LICENSE.txt
NOISE was created by Dan Staples (dismantl): http://disman.tl/ / noise@disman.tl.
The Markov code comes from https://code.google.com/p/kartoffelsalad/.
SOCKS proxy functionality from https://code.google.com/p/socksipy-branch/.
28 July 2013: Initial release.
30 July 2013: Added recurring dispatch schedule and more documentation.
4 August 2013: noise.py can now generate fake PGP-encrypted emails.
9 August 2013: Added STARTTLS, new config option, misc bug fix
13 August 2013: Performance improvements, misc bug fixes, added min-keywords option
17 August 2013: Major code refactoring, added Twitter dispatcher, misc fixes