AI Hub Dialogue Speech

Speech Dataset generator from AI Hub multimodal video dataset

Dataset Download

Request dataset from AI Hub Multimodal Video.

Convert to Speech Dataset

1. Place video dataset with following structure:

data
└── 0001-0400
    ├── clip_1
    │   ├── clip_1.json
    │   └── clip_1.mp4
    └── clip_2
        ├── clip_2.json
        └── clip_2.mp4

2. Install requirements

Requirements:

• moviepy==1.0.3
• librosa==0.8.0

3. Create audio file

python create_audio.py [--convert_video] [--sample_rate SR]

Option:

• convert_video: convert mp4 video into wav form first
• sample_rate: sampling rate (default: 22050)

4. Split train/dev/test set

python split.py [--path FILELIST_PATH] [--ratio RATIO] [--seed SEED]

Option:

• path: path of filelist. train/dev/test filelists are created in same directory.
• ratio: train_ratio:dev_ratio:test_ratio. Three ratios must be splitted with ':'
• seed: random seed for shuffling