Skip to content

vikigenius/neural_speaker_identification

Repository files navigation

Table of Contents

Description

  • ResNet based model for Speaker Identification
  • Evaluated on VoxCeleb1
  • Uses spectrograms as features

Prerequisites

See requirements.txt

Usage

Download VoxCeleb1 in data/raw. You should have a folder data/raw/VoxCeleb1

Build spectrograms and features using following command

python nsi.py featuregen --split='train' --show-pbar

Train the model

python nsi.py train