This is a reimplementation of the baseline code for our paper "See the silence: improving visual-only voice activity detection by optical flow and RGB fusion" available at: https://github.com/ducspe/VVADpaper
This baseline is compared to our new modified approach available at: https://github.com/ducspe/VisualOnlyVoiceActivityDetection.
The ResNet and LSTM architecture from the original authors was faithfully kept. The supporting infrastructure code around it was changed however to compress the original code significantly, since we do not need the audio branch and the audio-visual version present in the original implementation.
Unlike in the original implementation, we do not label manually, but rather implement a separate module that infers ground truth labels from the audio stream. Audio related code is available in the processing folder
The data_dda folder contains a subset of the preprocessed TCD-TIMIT dataset. The structure of this folder must be kept when working with the full dataset.
The preprocessed inputs can be obtained by using the code from: https://github.com/ducspe/TCD-TIMIT-Preprocessing
-
Notifications
You must be signed in to change notification settings - Fork 0
ducspe/ResNet18LSTMBaselineReimplementation
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
This is the baseline code to compare my new approach to.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published