corentinpla / hackathon-statup-2022

From a voice recording containing several voices and possibly music, we count the number of distincts speakers and the number of women and men who speak.

Team 2 PPTACL

• Paul Contamin https://www.linkedin.com/in/paul-contamin-70ab761b5/
• Antoine Joubrel https://www.linkedin.com/in/antoine-joubrel/
• Lucas Degeorge https://www.linkedin.com/in/lucas-degeorge-551a861a6/
• Corentin Pla https://www.linkedin.com/in/corentinpla/
• Thomas Dujardin https://www.linkedin.com/in/thomas-dujardin-8740bb207/

Main idea

From a voice recording containing several voices and possibly music, we count the number of distinct speakers and the number of women and men who speak.

Utilisation

The work has been split in two parts. It can be executed by using rendu_final.ipynb

Structure :

• Reconnaissance Femme_Homme allows the recognition of the speaker's gender from a voice recording with only one speaker and no music. Data used : LibriSpeech

• Separation_speacker in a soundtrack, allows to delimit the speeches and generates as many audio files as there are speakers

For optimal operation, the LibriSpeech database should be added to the folder "Recognition Woman_Man" in a folder named "LibriSpeech". The verification database of moviesoundclips.net program, should be added to the root of the repository, in a folder named "moviesoundclips.net_test".