BS2019

Hill Myna, a voice-only authentication system which aims to take into account spoofing attempts, such as replay attacks.

The main concept is based on an identification task, but it is augmented with a speech recognition task in order to defend its users from replay attacks. In fact, we rely on Microsoft Azure services both for the identification task, using the Azure Speaker Recognition service, and for the speech-to-text task, using the Azure Speech services.

The reason behind combining an identification task with a speech-to-text service is the following: provided the system is undergoing a replay attack, therefore the impostor has some kind of recording of our voice, without the speech-to-text service it might gain an easy access. Indeed, Hill Myna exploits a vocabulary of 100 words whenever a user tries to perform a login operation, by starting a challenge-response mechanism consisting in asking the user to repeat a random sample of words.

Since the words are randomly sampled from the vocabulary, it is highly unlikely that the impostor holds a voice sample containing those exact words, and therefore preventing the attack from being successful.

Instructions and requirements

Hill Myna has been developed on Ubuntu 18.04 (LTS) with Python 3.6+. In order to run this project, a Makefile has been set up to contain all the required libraries and Python packages; for this reason the suggested routine for running the project is the following:

1. Extract the 'hill_mina.zip' contents to a directory named 'hill_myna', in case of ZIP download; otherwise, clone this repository.
2. Move into the 'hill_myna' directory, containing 'Makefile', 'hill_myna.py' files, 'data', 'tmp' directories.
3. Run 'make' in this directory for installing all the required dependencies.
4. Run 'python hill_myna.py' and the GUI will be shown.
5. Follow instructions in the GUI.

Project structure

- Makefile                  # 'make' will take care of all the dependencies required
- hill_myna.py              # Starts the GUI
- backend                   # directory for all the backend source code
    |__ audio.py
    |__ azure.py
    |__ hill_myna_be.py
    |__ rest_client.py
    |__ users.py
    |__ words.py
- data                      # data needed by Hill Myna to run
    |__ credentials.csv     # Microsoft Azure credentials to access Speech Services and Cognitive Services
    |__ enrollment.txt      # Text users have to read for enrollment reasons
    |__ words.txt           # Words for anti-spoofing reasons
- tmp                       # directory for temporary files
- LICENSE                   # GNU AGPL v3.0
- README                    # this file

Funny fact

Hill myna is one of the many species of birds capable of imitating human speech. In this project, you will get to repeat random words like a typical hill myna would, in order to get identified and possibly avoid spoofing techniques such as replay attacks.