An API for voice based authentification, using GMM models and MFCC extraction, to create unique uservoice fingerprint. The model requires three audios (3-5 seconds) of a person for registration, and one audio for authentification. It exposes an API for 1 to 1 authentification, as well as 1 to N authentification. The audios sent on the api are automatically encoded for security purposes.
After installing Python 3 and pip3, Please launch the following bash file to install all requirements (it has only been tested on Linux systems):
sudo bash setup.sh
Please also install mongodb on your machine, and create a database with the name identification, and a collection with the name users
You can launch the server on the terminal with :
python3 app.py
-
To create the enrollement input file
enrollement.json:python3 create_enrollement_input.py -a voice_database/ihab/recording.wav -b voice_database/unknown/flute.wav -c voice_database/ihab/recording.wav -
To create the 1-to-1 authentification input file
auth.json:python3 create_auth_input.py -a voice_database/unknown/flute.wav -b gmm_models/ihab.gmm -
To create the 1-to-N authentification input file
auth.json:python3 create_auth_input.py -a voice_database/unknown/flute.wav
All API responses are the same as specified in the specifications
- For the main enrollment, this is an example of the curl request to send :
curl -X POST -F audio_file=@enrollement.json 'http://127.0.0.1:5000/api/voiceident/enrollments?username=ihab'
You can specify the username you desire, as well as the path for the audio file you would like to use.
-
For checking advancement of enrollement, this is an example of the curl request necessary :
curl -i -X GET 'http://127.0.0.1:5000/api/voiceident/enrollments?username=ihab&voiceprintId=5df8cb542e276886223cbe36'The voiceprintid is returned in the first API calls -
For authentification, the curl request is :
curl -X POST -F audio_file=@auth.json 'http://127.0.0.1:5000/api/voiceident/authentication?username=ihab&voiceprintId=5df8cb542e276886223cbe36'
