Audio List Processing for Speaker Verification

Overview

Preparing audio list for Speaker Verificaton in Pytorch using kaldi and python scripts. Most of scripts are borrowed from Kaldi.

Dataset

VoxCeleb1&2, SITW, Aishell, Hi-Mia, ...

Directorys

├── root
│   ├── conf            # Config files for features, etc.
│   ├── local           # Preparing wav.scp, utt2spk, etc.
│   ├── prepare_data    # Making features in dataset dirs.
│   ├── sid             # copied from kaldi
│   ├── steps           #  ...
│   ├── utils           #  ...

Verification Trials Generation

python local/make_trials.py --data-dir <dir_with_wav.scp> --num-pair <num_of_pairs> --ignore-gender <cross_gender> --gender <specific_gender> --trials <trials_name>