vggvox

Python adaptation of VGGVox speaker identification model, based on Nagrani et al 2017, "VoxCeleb: a large-scale speaker identification dataset"
Evaluation code only, based on the author's Matlab code and pretrained model.

Instructions

Install python3 and the required packages
Modify cfg/enroll_list.csv and cfg/test_list.csv to point to your local enroll/test wav files
To run evaluation: python3 scoring.py
Results will be stored in res/results.csv. Each line has format: [path to test wav], [correct speaker], [distance to enroll speaker 1],...[distance to enroll speaker N], [predicted speaker], [correct?]

Name		Name	Last commit message	Last commit date
Latest commit History 8 Commits
cfg		cfg
data		data
.gitignore		.gitignore
README.md		README.md
constants.py		constants.py
model.py		model.py
scoring.py		scoring.py
sigproc.py		sigproc.py
wav_reader.py		wav_reader.py