synthesis.py

import time
import torch
import torch.nn.functional as F
import os
import sys
import numpy as np
import torch.nn as nn
import argparse

import src
from utils.utilities import (save_json, compute_time, print_dict, mkdir)
from inference.inference import Inference
from inference.compute_measure import (evaluate_transcription, evaluate_separation)


if __name__=='__main__':

	parser = argparse.ArgumentParser(description='')
	parser.add_argument('--model_name', type=str, required=True, help='Model name in \
																																					[`AMT` for trainscription-only baseline, \
																																						`MSS` for separation-only baseline, \
																																						`MSS-AMT` for multi-task baseline, \
																																						`MSI` for the proposed multi-task score-informed model, \
																																						`MSI-DIS` for the proposed multi-task score-informed with further disentanglement model].')
	parser.add_argument('--model_path', type=str, required=True, help='Model weights path.')
	parser.add_argument('--evaluation_folder', type=str, required=True, help='Directory to store evaluation results.')
	parser.add_argument('--epoch', type=str, required=True, help='Epoch.')


	args = parser.parse_args()

	model_name = args.model_name
	model_path = args.model_path
	output_dir = args.evaluation_folder
	epoch = args.epoch

	evaluation_dir = f"{output_dir}/{model_name}"

	path = "songs/road.mid"

	inference = Inference(model_name, model_path, evaluation_dir, epoch)
	inference.synthesis(path, 0)