infer_cli_edna.py

from pathlib import Path
import os
import json
import argparse

from dotenv import load_dotenv
from scipy.io import wavfile

from infer_script.vc_script.modules import VC

load_dotenv()

parser = argparse.ArgumentParser(
                    prog='Conversion Pipeline',
                    description='Runs conversion pipeline for RVC',
                    epilog='Ask me for help')
parser.add_argument('parfile') 

def read_param_dict(parser):
    args = parser.parse_args()
    json_file = open(args.parfile)
    param_dict = json.load(json_file)
    json_file.close()

    return param_dict

rvc_dict = read_param_dict(parser)

root = '/media/HDD_disk/tomas/ICHOIR/fork/Retrieval-based-Voice-Conversion-WebUI'

model_name = rvc_dict["model_name"]
index_file = rvc_dict["index_file"]

model_path = f'{model_name}.pth' 
index_path = f'{root}/logs/{model_name}/{index_file}'
hubert_path = f'{root}/assets/hubert/{rvc_dict["hubert_file"]}'
input_path = rvc_dict["input_path"]

wav_name = input_path.split('/')[-1].split('.wav')[-2]
f0_method = rvc_dict["f0_method"]
protect = rvc_dict["protect"]
f0_up_key = rvc_dict["f0_up_key"]

experiment_name = f'{wav_name}_by_{model_name}_f0_method_{f0_method}_protect_{protect}_f0_up_key_{f0_up_key}'
experiment_dir = f'{root}/audio_rvc_output/{experiment_name}'

os.mkdir(experiment_dir)
output_path = f'{experiment_dir}/{experiment_name}.wav'

with open(f"{experiment_dir}/metadata.json", "w") as outfile: 
    json.dump(rvc_dict, outfile)

vc = VC()
vc.get_vc(model_path, index_file = index_path)


print('--------------- Processing audio')
tgt_sr, audio_opt, times, _ = vc.vc_inference(1, input_path, #Path(input_audio),
                                                hubert_path = hubert_path,
                                                f0_method = f0_method,
                                                f0_up_key = f0_up_key,
                                                protect = protect)

print(f'--------------- Saving audio to {output_path}')
wavfile.write(output_path, tgt_sr, audio_opt)