Skip to content

Light CNN for ASVSpoof 2019 implemented on Tensorflow and Keras

License

Notifications You must be signed in to change notification settings

ozora-ogino/LCNN

Repository files navigation

Light CNN for ASVSpoof (Tensorflow-Keras)

Test

Description

Light CNN (LCNN) is CNN based model which was proposed in Interspeech 2019 by STC teams and state of the art of ASVspoof2019.

LCNN is featured by max feature mapping function (MFM). MFM is an alternative of ReLU to suppress low-activation neurons in each layer. MFM contribute to make LCNN lighter and more efficient than CNN with ReLU.

If you'd like to know more detail, see the references below.

Experiment setup

In this project, LCNN is trained with ASVspoof2019 PA dataset. As a speech feature, I used spectrograms that extracted by using STFT or CQT.

Reference

"A Light CNN for Deep Face Representation with Noisy Labels"

"STC Antispoofing Systems for the ASVspoof2019 Challenge"

ASVspoof2019

Contributing

Interested in contributing? Awesome! Fork and create PR! Or you can post Issue for bug reports or your requests (e.g. pytorch support).

About

Light CNN for ASVSpoof 2019 implemented on Tensorflow and Keras

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published