Voice Re-identification

Voice re-id is an application that allows a person to be recognized by their voice.

How It Works

There are two directories PCA and no_PCA Both PCA and no_PCA perform reidentification from the speech signal via opened_set and closed_set and verification. In the opened_set system it is possible also to run "run-time-reid" and be re-identified after registering in the system. Accuracy values ​​are reported in the output directory inside the current directory. Inside no_PCA there are two directories single_training and training_rinforzato, they are two different ways to do training.

Running Instructions

Windows

Linux

For PCA/verifica, inside this directory run:

bash start.sh

and the output are presents in the directory PCA/verifica/output

For PCA/identification_close_set, inside this directory run:

bash start.sh

and the output are presents in the directory PCA/identification_close_set/output

For no_PCA/single_training/identification_closed_set, inside this directory run:

bash start.sh

and the output are presents in the directory no_PCA/single_training/identification_closed_set/output

For no_PCA/single_training/identification_open_set, inside this directory run:

bash start.sh

and the output are presents in the directory no_PCA/single_training/identification_open_set/output

For no_PCA/single_training/verifica, inside this directory run:

bash start.sh

and the output are presents in the directory no_PCA/single_training/verifica/output

For no_PCA/single_training/runtimereid, inside this directory run:

bash reid_runtime.sh

For no_PCA/5-cross-validation is the same!!

Installation

You can use the requirements.txt file to install packages with pip (created with the command pip freeze > requirements.txt):

pip3 install -r requirements.txt

Dataset

The dataset is downloaded from the link.

Output Example

This is the Confusion Matrix with 5 cross validation in Verification, without PCA