Entry point

• dpo_claps_asr.ipynb

• related files: dpo_eval.py, vc/, CLAPS/

Data Source

soxdata_encodec

Use TMUX to run the script in the background

sudo apt-get update && sudo apt-get install tmux
pip install nbconvert
jupyter nbconvert --to script dpo_claps_asr.ipynb.ipynb

(alternatively, in VSCode, press ctrl + shift + p and click Jupyter: Export to Python Script)

Use plot.ipynb to plot the result

• be sure to modify the model_output_dir (it is ok to use tensorboard to visualize the result, however, the plot.ipynb is more flexible and better for report)

Note

• Reward function is of the form reward = k*(CLAPS_reward) + (1-k)*(ASR_reward), where k is a hyperparameter.

Tensorboard Usage

tensorboard --logdir="model_output/1210-2215/tensorboard_logs"