/AAI2022Fall-Project

AAI Project: Text-Independent Speaker Identification

Primary LanguageJupyter NotebookMIT LicenseMIT

AAI2022Fall-Project

AAI Project: Text-Independent Speaker Identification

Warning: This repo is only for clean voices, which do not include noises.

Main Dependencies

  • python 3.8
  • torch 1.12.1
  • torchaudio 0.12.1
  • torchvision 0.13.1

Forwarding Pipeline

.flac -> torchaudio.load -> tensor
tensor -> silding window -> n tensor fragments, same length
fragments -> MFCC transformation -> n spectral tensors, size: H * W
spectral tensors -> CNN models: MFCCNN or ResNet18 -> n predicted labels
predicted labels -> most frequent label -> y_pred

MFCCNN: H=W=64; ResNet18: H=W=224

Usage

cd model-dz/

# Train
python -u train.py -m mfccnn -g <GPUID> > mfccnn.log 2>&1
python -u train.py -m resnet -g <GPUID> > resnet.log 2>&1

# MFCCNN 0.96 test acc
# ResNet 0.99 test acc

# Inference
python train.py -m mfccnn/resnet -g <GPUID> -f <file_path>.flac
python train.py -m mfccnn/resnet -g <GPUID> # all test files

说明

完整项目链接: https://github.com/Cui-Inc-Lab/AAI2022Fall-Project

此仓库只包含所有我自己独立完成的部分，因为我实在受不了那个管理混乱的原项目了

永远不要尝试教别人 git 项目管理，我破防了

bef0e4258b4208be5149c9b5f884a12b9