vad

Voice Activity Detection

To run an VAD example, execute the following commands from your Athena root directory:

source env.sh
bash examples/vad/$dataset_name/run.sh

Core Stages:

(1) Data download

First, the coorsponding dataset should be downloaded and stored in examples/vad/$dataset_name/data/resource. In google_dataset_v2 example, both speech commands in Google Speech Commands Dataset V2 dataset and noises in openslr-28 database are used.

(2) Data preparation

By using examples/vad/$dataset_name/local/prepare_data.sh, the dataset would be divided into three parts separately as training set, validation set and testing set. After that, the features of all of them will be extracted and stored in kaldiio format.

(3) Model training

Two kinds of VAD models, DNN and MarbleNet, are supported in Athena. You can using json file examples/vad/$dataset_name/configs/vad_dnn.json and examples/vad/$dataset_name/vad_marbleNet.json to performe the training of them respectively. Both of DNN and MarbleNet are frame-level classifiers. The prediction resulsts is generated with overlapping input segments, whose length is specified by left_context and right_context in json files.

(4) Inferencing

Currently, you can performe frame-level inferencing. To do it, run

$ python athena/inference.py examples/vad/google_dataset_v2/configs/vad_$model.json

Task	Model Name	Training Data	Input Segment	Frame Error Rate
VAD	DNN	Google Speech Commands Dataset V2	0.21s	8.49%
VAD	MarbleNet	Google Speech Commands Dataset V2	0.63s	2.50%