Simplify CLI

[hbredin]

The whole {root_dir}/train/{protocol}.{subset}/apply/{date}/{protocol}.{subset}.rttm directory scheme is (a bit) cumbersome.

Would be nice to simplify the whole thing:

Training

pyannote-pipeline train {/path/to/config.yml} {protocol}

would creates two files in the same directory as config.yml:

• /path/to/{protocol}.{subset}.yml that contains everything needed to load the trained pipeline

  from pyannote.audio import Pipeline
  pipeline = Pipeline.from_pretrained("/path/to/{protocol}.{subset}.yml")

• /path/to/{protocol}.{subset}.db that contains optuna experiments

  optuna-dashboard sqlite:////path/to{protocol}.{subset}.db

Inference

pyannote-pipeline apply /path/to/{protocol}.{subset}.yml {protocol} 
pyannote-pipeline apply /path/to/{protocol}.{subset}.yml /path/to/audio.wav 

[hbredin]

Also, for {protocol} inference, it would be nice to have some way of keeping track of the hyper-parameter used directly in the resulting RTTM file.