Theo Lepage

icon indicating a website link https://theolepage.com icon indicating an email [email protected]

icon location Paris, France icon location Ph.D. student in A.I. → Self-Supervised Learning for Speaker Recognition

Results 3 comments of Theo Lepage

ThinResNet34

Hello, Indeed, the implementation of the Thin ResNet-34 encoder from "Experimenting with Additive Margins for Contrastive Self-Supervised Speaker Verification" is different from the one used in "Label-Efficient Self-Supervised Speaker Verification...

How does SSPS evaluate on VoxCeleb1-O dataset?

Hi, During training (using the `train*.py` scripts), evaluation is performed on 4-second segments (64000 samples) to speed up computation. In contrast, full-length utterances are used during evaluation (with the `evaluate*.py`...

Example Application Source Code

It seems that the example app is available here: [https://github.com/YZarytskyi/react-audio-visualization](https://github.com/YZarytskyi/react-audio-visualization)