gan_test.py

import os
import glob
import torch
import torchvision.utils as vutils
import webrtcvad

from mfcc import MFCC
from config import DATASET_PARAMETERS, NETWORKS_PARAMETERS
from network import get_network
from utils import voice2face

# initialization
vad_obj = webrtcvad.Vad(2)
mfc_obj = MFCC(nfilt=64, lowerf=20., upperf=7200., samprate=16000, nfft=1024, wlen=0.025)
e_net, _ = get_network('e', NETWORKS_PARAMETERS, train=False)
g_net, _ = get_network('g', NETWORKS_PARAMETERS, train=False)

# test
voice_path = os.path.join(DATASET_PARAMETERS['test_data'], '*.wav')
voice_list = glob.glob(voice_path)
for filename in voice_list:
    face_image = voice2face(e_net, g_net, filename, vad_obj, mfc_obj,
                            NETWORKS_PARAMETERS['GPU'])
    vutils.save_image(face_image.detach().clamp(-1,1),
                      filename.replace('.wav', '.png'), normalize=True)