search

SpectData / MONAH

Multi-Modal Narratives for Humans
MIT License
3 stars 0 forks source link

Creation of verbatim transcript #5

Closed joshkyh closed 4 years ago

joshkyh commented 4 years ago

(1) The demo files, have split videos after removing the joint audio. However, when the separateed audio is added to the invidual videos, the videos seemed corrupted. (2) The PyLint has been started. Completed 1 file.

Proritize (1). PyLint work can be shared or deprioritized.

joshkyh commented 4 years ago

Using a online tool, you have managed to mix in the audio for each video. Great.

Next Steps: (1) Aim for next Tue end-to-end vpa creation. I.e., given two videos, create the vpa script upon run. (2) Some small minority of pylint conventions can't be removed, when that happens -- use pylint suppressions, declared at the top of the .py file.

joshkyh commented 4 years ago

Thursday, demonstration of entire vpa. Well done @mfmakahiya

joshkyh commented 4 years ago

(1) Prepare a configuration dictionary to be passed into the various second layer methods. (2) Second layer refers to convert the output of multimodal information extraction (1st layer) into the vpa transcript. Have a separate folder for the second layer scripts. (3) Have a naming convention for all confuration keys, so that it is easy to organize your configuration keys. (4) We decided that each layers' manager will be in the same layer. stage1_manager, stage2_manager, stage3_manager. stage3_manager will call stage1 and stage2. stage1 and stage2 will orchestrate the calling of the various python scripts. (low-level)

joshkyh commented 4 years ago

(1) Organize the outputs for csv for video1name_video2name > Stage_1, Stage_2, Stage_3 subfolders (2) Add speaker names in config.py. Then add video1name and video2name when you call the main function (3) Stage 1 Verbatim should create a word level timing csv, so that people can bring their own csv

joshkyh commented 4 years ago

Great work. vpa weaved. No follow up actions.