def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
avesdr = 0
numSongs = 0
sdrmedian = np.zeros(50)
for iloader, xtrain, ytrain in loadtest:
iloader=iloader.item()
listofpred0 = []
cnt,aveloss=0,0
for ind in range(0, xtrain.shape[-1] - sampleSize, sampleSize):
if (xtrain[0, 0, ind:ind + sampleSize].shape[0] < (sampleSize)): break
output = model(xtrain[:, :,ind:ind + sampleSize].to(device))
listofpred0.append(output.reshape(-1).cpu().numpy())
loss = criterion(output, (ytrain[:, :,ind:ind + sampleSize].to(device)))
cnt+=1
aveloss += float(loss)
aveloss /= cnt
print('loss for test:{},num{},epoch{}'.format(aveloss, iloader,epoch))
ans0 = mu_law_decode(np.concatenate(listofpred0))
if(iloader >= 150):
sdr = bss_eval_sources(mu_law_decode(ytrain[0,0,:ans0.shape[0]].cpu().numpy()), ans0)
avesdr += sdr[0][0]
sdrmedian[iloader-150] = sdr[0][0]
#print('each ele of median',sdrmedian[iloader-150],iloader-150)
numSongs += 1
if(iloader > 160):continue
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
sf.write(savemusic.format(iloader), ans0, sample_rate)
print('test stored done', np.round(time.time() - start_time))
print('sdr mean:', avesdr / numSongs)
print('sdr median:', np.median(sdrmedian))
def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
for iloader, y, queue in loadval:
iloader = iloader.item()
Fs = 16000
f = 500
sample = 16000 * 7
x = np.arange(sample)
ytrain = np.sin(2 * np.pi * f * x / Fs)
ytrain = mu_law_encode(ytrain)
y = torch.from_numpy(ytrain.reshape(1, -1)).type(torch.LongTensor)
#music = model.infer(queue,device)
music = model.slowInfer(queue, device, y, 16000 * 0.001)
music = model.slowInfer(queue, device, y, 16000 * 0.01)
music = model.slowInfer(queue, device, y, 16000 * 0.1)
#music = model.slowInfer(queue, device, y, 16000 * 1)
music = model.slowInfer(queue, device, y, 16000 * 3)
print(music[:100])
ans0 = mu_law_decode(music.numpy().astype('int'))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
write(savemusic.format(iloader), sample_rate, ans0)
print('test stored done', np.round(time.time() - start_time))
def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
for iloader, xtrain, ytrain in loadval:
iloader = iloader.item()
listofpred0 = []
cnt, aveloss = 0, 0
for ind in range(pad, xtrain.shape[-1] - pad - sampleSize,
sampleSize):
output0 = model(xtrain[:, :, ind - pad:ind + sampleSize +
pad].to(device))
loss = criterion(output0,
(ytrain[:, ind:ind + sampleSize].to(device)))
pred0 = output0.max(1,
keepdim=True)[1].cpu().numpy().reshape(-1)
listofpred0.append(pred0)
cnt += 1
aveloss += loss.item()
aveloss /= cnt
print('loss for validation:{},num{},epoch{}'.format(
aveloss, iloader, epoch))
#if(aveloss < maxloss[iloader]):
maxloss[iloader] = aveloss
ans0 = mu_law_decode(np.concatenate(listofpred0))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
sf.write(savemusic.format(iloader), ans0, sample_rate)
print('test stored done', np.round(time.time() - start_time))
def test(iloader, xtrain): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
listofpred0, listofpred1 = [], []
for ind in range(pad, xtrain.shape[-1] - pad, sampleSize):
output0, output1 = model(xtrain[:, :, ind - pad:ind + sampleSize +
pad].to(device))
pred0 = output0.max(1, keepdim=True)[1].cpu().numpy().reshape(-1)
pred1 = output1.max(1, keepdim=True)[1].cpu().numpy().reshape(-1)
listofpred0.append(pred0)
listofpred1.append(pred1)
ans0 = mu_law_decode(np.concatenate(listofpred0))
ans1 = mu_law_decode(np.concatenate(listofpred1))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
sf.write(savemusic0.format(iloader), ans0, sample_rate)
sf.write(savemusic1.format(iloader), ans1, sample_rate)
print('test stored done', np.round(time.time() - start_time))
def test(xtrain,iloader): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
#for iloader, (xtest, _) in enumerate(loadval):
listofpred = []
for ind in range(pad, xtrain.shape[-1] - pad, sampleSize):
output = model(xtrain[:, :, ind - pad:ind + sampleSize + pad].to(device))
pred = output.max(1, keepdim=True)[1].cpu().numpy().reshape(-1)
listofpred.append(pred)
ans = mu_law_decode(np.concatenate(listofpred))
sf.write(savemusic.format(iloader), ans, sample_rate)
print('test stored done', time.time() - start_time)
def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
for iloader, xtrain in loadtest:
iloader = iloader.item()
print(iloader)
listofpred0 = []
for ind in range(0, xtrain.shape[-1] - sampleSize, sampleSize):
output = model(xtrain[:, :, ind:ind + sampleSize].to(device))
listofpred0.append(output.reshape(-1))
ans0 = mu_law_decode(np.concatenate(listofpred0))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
sf.write(savemusic.format(iloader), ans0, sample_rate)
print('test stored done', np.round(time.time() - start_time))
def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
for iloader, y in loadval:
iloader = iloader.item()
#music = model.infer(queue)
#music = model.slowInfer(queue, y, 16000 * 0.001)
#music = model.slowInfer(queue, y, 16000 * 0.01)
#music = model.slowInfer(queue, y, 16000 * 0.1)
music = model.slowInfer(None, y, 16000 * 3)
print(music[:1000])
ans0 = mu_law_decode(music.numpy().astype('int'))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
write(savemusic.format(epoch * 6 + iloader), sample_rate, ans0)
print('test stored done', np.round(time.time() - start_time))
def test(epoch): # testing data
model.eval()
start_time = time.time()
with torch.no_grad():
for iloader, xtrain, ytrain in loadtest:
iloader = iloader.item()
listofpred0 = []
cnt, aveloss = 0, 0
for ind in range(0, xtrain.shape[-1] - sampleSize, sampleSize):
output = model(xtrain[:, :, ind:ind + sampleSize].to(device))
listofpred0.append(output.reshape(-1))
loss = criterion(
output, (ytrain[:, :, ind:ind + sampleSize].to(device)))
cnt += 1
aveloss += float(loss)
aveloss /= cnt
print('loss for test:{},num{},epoch{}'.format(
aveloss, iloader, epoch))
ans0 = mu_law_decode(np.concatenate(listofpred0))
if not os.path.exists('vsCorpus/'): os.makedirs('vsCorpus/')
sf.write(savemusic.format(iloader), ans0, sample_rate)
print('test stored done', np.round(time.time() - start_time))