Skip to content

Latest commit

 

History

History
34 lines (22 loc) · 589 Bytes

README.md

File metadata and controls

34 lines (22 loc) · 589 Bytes
Raw

Data Preprocess

Before train or predict, we need a pre-process to convert the data into feature vectors.

Generating training data

AVSpeech

  1. Put mp4 files in AVSPEECH_DIR.

  2. Run (GPU Only)

$ docker-compose run preprocess ./run.sh

AudioSet

  1. Put mp4 files in AUDIOSET_DIR.

  2. Run

$ docker-compose run preprocess python3 convert_audioset.py

Generating predicting data

  1. Set mode = Mode.predict appeared in preprocess/src/env.py.

  2. Put mp4 files in MOVIE_DIR.

  3. Run (GPU Only)

$ docker-compose run preprocess ./run.sh