Implementation of [1706.08612] VoxCeleb: a large-scale speaker identification dataset
Speaker Recognition Scores:
| Metric | Paper | Ours |
|---|---|---|
| Top-5 | 92 | 90.7 |
| Top-1 | 81 | 78 |
Modifications:
- Added a Local cepstral mean and variance normalization over a sliding window of 3s
TODO:
- Add VAD to capture speech frames only