SSL-PVAD

A repository for code used to produce the results the ICASSP 2024 paper: "SELF-SUPERVISED PRETRAINING FOR ROBUST PERSONALIZED VOICE ACTIVITY DETECTION IN ADVERSE CONDITIONS"

Installing packages used in project

The packages used in this project can be installed by running:

pip install -r requirements.txt

Data preparation

To create the data sets used in this study run:

python data_preprocessing/prepare_data_librispeech_concat.py 
--conf configs/data_preparation/librispeech_concat_config.yaml
--generate_utterances
--generate_embeddings
--unique 

Here, you will need to update the configuration file with your data paths.

The script automatically downloads the LibriSpeech data used to generate the multi-speaker utterances, so make sure you have enough disk space before you run it.

Running experiments

To run the experiments, simply run

python <train_script_name> --conf <path_to_config>

E.g.,

python train_APC --conf configs/APC/LSTM_D64L2_APC_pretrain_960h_config_100_epoch.yaml

Examples of configs are found in the configsfolder.