def main(conf):
# Make the model
model, _ = make_model_and_optimizer(conf['train_conf'])
# Load best model
with open(os.path.join(conf['exp_dir'], 'best_k_models.json'), "r") as f:
best_k = json.load(f)
best_model_path = min(best_k, key=best_k.get)
# Load checkpoint
checkpoint = torch.load(best_model_path, map_location='cpu')
state = checkpoint['state_dict']
state_copy = state.copy()
# Remove unwanted keys
for keys, values in state.items():
if keys.startswith('loss'):
del state_copy[keys]
print(keys)
model = torch_utils.load_state_dict_in(state_copy, model)
# Handle device placement
if conf['use_gpu']:
model.cuda()
model_device = next(model.parameterss()).device
test_set = LibriMix(csv_dir=conf['test_dir'],
task=conf['task'],
sample_rate=conf['sample_rate'],
n_src=conf['train_conf']['data']['n_src'],
segment=None) # Uses all segment length
# Used to reorder sources only
loss_func = PITLossWrapper(pairwise_neg_sisdr, pit_from='pw_mtx')
# Randomly choose the indexes of sentences to save.
eval_save_dir = os.path.join(conf['exp_dir'], conf['out_dir'])
ex_save_dir = os.path.join(eval_save_dir, 'examples/')
if conf['n_save_ex'] == -1:
conf['n_save_ex'] = len(test_set)
save_idx = random.sample(range(len(test_set)), conf['n_save_ex'])
series_list = []
torch.no_grad().__enter__()
for idx in tqdm(range(len(test_set))):
# Forward the network on the mixture.
mix, sources = tensors_to_device(test_set[idx], device=model_device)
est_sources = model(mix.unsqueeze(0))
loss, reordered_sources = loss_func(est_sources,
sources[None],
return_est=True)
mix_np = mix.cpu().data.numpy()
sources_np = sources.squeeze().cpu().data.numpy()
est_sources_np = reordered_sources.squeeze().cpu().data.numpy()
# For each utterance, we get a dictionary with the mixture path,
# the input and output metrics
utt_metrics = get_metrics(mix_np,
sources_np,
est_sources_np,
sample_rate=conf['sample_rate'])
utt_metrics['mix_path'] = test_set.mixture_path
series_list.append(pd.Series(utt_metrics))
# Save some examples in a folder. Wav files and metrics as text.
if idx in save_idx:
local_save_dir = os.path.join(ex_save_dir, 'ex_{}/'.format(idx))
os.makedirs(local_save_dir, exist_ok=True)
sf.write(local_save_dir + "mixture.wav", mix_np,
conf['sample_rate'])
# Loop over the sources and estimates
for src_idx, src in enumerate(sources_np):
sf.write(local_save_dir + "s{}.wav".format(src_idx), src,
conf['sample_rate'])
for src_idx, est_src in enumerate(est_sources_np):
sf.write(local_save_dir + "s{}_estimate.wav".format(src_idx),
est_src, conf['sample_rate'])
# Write local metrics to the example folder.
with open(local_save_dir + 'metrics.json', 'w') as f:
json.dump(utt_metrics, f, indent=0)
# Save all metrics to the experiment folder.
all_metrics_df = pd.DataFrame(series_list)
all_metrics_df.to_csv(os.path.join(eval_save_dir, 'all_metrics.csv'))
# Print and save summary metrics
final_results = {}
for metric_name in compute_metrics:
input_metric_name = 'input_' + metric_name
ldf = all_metrics_df[metric_name] - all_metrics_df[input_metric_name]
final_results[metric_name] = all_metrics_df[metric_name].mean()
final_results[metric_name + '_imp'] = ldf.mean()
print('Overall metrics :')
pprint(final_results)
with open(os.path.join(eval_save_dir, 'final_metrics.json'), 'w') as f:
json.dump(final_results, f, indent=0)
def main(conf):
compute_metrics = update_compute_metrics(conf["compute_wer"],
COMPUTE_METRICS)
anno_df = pd.read_csv(
Path(conf["test_dir"]).parent.parent.parent / "test_annotations.csv")
wer_tracker = (MockWERTracker() if not conf["compute_wer"] else WERTracker(
ASR_MODEL_PATH, anno_df))
model_path = os.path.join(conf["exp_dir"], "best_model.pth")
model = DPRNNTasNet.from_pretrained(model_path)
# Handle device placement
if conf["use_gpu"]:
model.cuda()
model_device = next(model.parameters()).device
test_set = LibriMix(
csv_dir=conf["test_dir"],
task=conf["task"],
sample_rate=conf["sample_rate"],
n_src=conf["train_conf"]["data"]["n_src"],
segment=None,
return_id=True,
) # Uses all segment length
# Used to reorder sources only
loss_func = PITLossWrapper(pairwise_neg_sisdr, pit_from="pw_mtx")
# Randomly choose the indexes of sentences to save.
eval_save_dir = os.path.join(conf["exp_dir"], conf["out_dir"])
ex_save_dir = os.path.join(eval_save_dir, "examples/")
if conf["n_save_ex"] == -1:
conf["n_save_ex"] = len(test_set)
save_idx = random.sample(range(len(test_set)), conf["n_save_ex"])
series_list = []
torch.no_grad().__enter__()
for idx in tqdm(range(len(test_set))):
# Forward the network on the mixture.
mix, sources, ids = test_set[idx]
mix, sources = tensors_to_device([mix, sources], device=model_device)
est_sources = model(mix.unsqueeze(0))
loss, reordered_sources = loss_func(est_sources,
sources[None],
return_est=True)
mix_np = mix.cpu().data.numpy()
sources_np = sources.cpu().data.numpy()
est_sources_np = reordered_sources.squeeze(0).cpu().data.numpy()
# For each utterance, we get a dictionary with the mixture path,
# the input and output metrics
utt_metrics = get_metrics(
mix_np,
sources_np,
est_sources_np,
sample_rate=conf["sample_rate"],
metrics_list=COMPUTE_METRICS,
)
utt_metrics["mix_path"] = test_set.mixture_path
est_sources_np_normalized = normalize_estimates(est_sources_np, mix_np)
utt_metrics.update(**wer_tracker(
mix=mix_np,
clean=sources_np,
estimate=est_sources_np_normalized,
wav_id=ids,
sample_rate=conf["sample_rate"],
))
series_list.append(pd.Series(utt_metrics))
# Save some examples in a folder. Wav files and metrics as text.
if idx in save_idx:
local_save_dir = os.path.join(ex_save_dir, "ex_{}/".format(idx))
os.makedirs(local_save_dir, exist_ok=True)
sf.write(local_save_dir + "mixture.wav", mix_np,
conf["sample_rate"])
# Loop over the sources and estimates
for src_idx, src in enumerate(sources_np):
sf.write(local_save_dir + "s{}.wav".format(src_idx), src,
conf["sample_rate"])
for src_idx, est_src in enumerate(est_sources_np_normalized):
sf.write(
local_save_dir + "s{}_estimate.wav".format(src_idx),
est_src,
conf["sample_rate"],
)
# Write local metrics to the example folder.
with open(local_save_dir + "metrics.json", "w") as f:
json.dump(utt_metrics, f, indent=0)
# Save all metrics to the experiment folder.
all_metrics_df = pd.DataFrame(series_list)
all_metrics_df.to_csv(os.path.join(eval_save_dir, "all_metrics.csv"))
# Print and save summary metrics
final_results = {}
for metric_name in compute_metrics:
input_metric_name = "input_" + metric_name
ldf = all_metrics_df[metric_name] - all_metrics_df[input_metric_name]
final_results[metric_name] = all_metrics_df[metric_name].mean()
final_results[metric_name + "_imp"] = ldf.mean()
print("Overall metrics :")
pprint(final_results)
if conf["compute_wer"]:
print("\nWER report")
wer_card = wer_tracker.final_report_as_markdown()
print(wer_card)
# Save the report
with open(os.path.join(eval_save_dir, "final_wer.md"), "w") as f:
f.write(wer_card)
with open(os.path.join(eval_save_dir, "final_metrics.json"), "w") as f:
json.dump(final_results, f, indent=0)
model_dict = torch.load(model_path, map_location="cpu")
os.makedirs(os.path.join(conf["exp_dir"], "publish_dir"), exist_ok=True)
publishable = save_publishable(
os.path.join(conf["exp_dir"], "publish_dir"),
model_dict,
metrics=final_results,
train_conf=train_conf,
)
def main(conf):
model_path = os.path.join(conf["exp_dir"], "best_model.pth")
model = TransMask.from_pretrained(model_path)
# Handle device placement
if conf["use_gpu"]:
model.cuda()
model_device = next(model.parameters()).device
if conf['file_path'] == '':
test_set = LibriMix(
csv_dir=conf["test_dir"],
task=conf["task"],
sample_rate=conf["sample_rate"],
n_src=conf["train_conf"]["masknet"]["n_src"],
segment=None,
) # Uses all segment length
# Used to reorder sources only
loss_func = PITLossWrapper(pairwise_neg_sisdr, pit_from="pw_mtx")
# Randomly choose the indexes of sentences to save.
eval_save_dir = os.path.join(conf["exp_dir"], conf["out_dir"])
ex_save_dir = os.path.join(eval_save_dir, "examples/")
if conf["n_save_ex"] == -1 and conf['file_path'] == '':
conf["n_save_ex"] = len(test_set)
save_idx = random.sample(range(len(test_set)), conf["n_save_ex"])
else:
save_idx = 0
series_list = []
torch.no_grad().__enter__()
sdr = 0
rtf = 0
if conf['file_path'] != '':
file_path = conf['file_path']
if os.path.isdir(file_path):
wavs = [
os.path.join(file_path, wav) for wav in os.listdir(file_path)
if '.wav' in wav
]
for wav in wavs:
inference_wav(wav, conf, model_device, model, ex_save_dir)
else:
inference_wav(file_path, conf, model_device, model, ex_save_dir)
return
for idx in tqdm(range(len(test_set))):
# Forward the network on the mixture.
mix, sources = tensors_to_device(test_set[idx], device=model_device)
mul = 8
mix = mix.view(-1, 1).repeat(1, mul).view(-1)
sources = sources.repeat(1, mul)
#print('DEVICE')
#print(model_device)
ss = time()
est_sources = model(mix.unsqueeze(0))
dur = time() - ss
ll = len(mix) / 8000
rtf += (dur / ll)
print(rtf / (idx + 1))
#import pdb;pdb.set_trace()
loss, reordered_sources = loss_func(est_sources,
sources[None],
return_est=True)
mix_np = mix.cpu().data.numpy()
sources_np = sources.cpu().data.numpy()
est_sources_np = reordered_sources.squeeze(0).cpu().data.numpy()
# For each utterance, we get a dictionary with the mixture path,
# the input and output metrics
utt_metrics = get_metrics(
mix_np,
sources_np,
est_sources_np,
sample_rate=conf["sample_rate"],
metrics_list=compute_metrics,
)
sdr += utt_metrics['sdr']
print(sdr / (idx + 1))
utt_metrics["mix_path"] = test_set.mixture_path
series_list.append(pd.Series(utt_metrics))
# Save some examples in a folder. Wav files and metrics as text.
if idx in save_idx:
local_save_dir = os.path.join(ex_save_dir, "ex_{}/".format(idx))
os.makedirs(local_save_dir, exist_ok=True)
sf.write(local_save_dir + "mixture.wav", mix_np,
conf["sample_rate"])
# Loop over the sources and estimates
for src_idx, src in enumerate(sources_np):
sf.write(local_save_dir + "s{}.wav".format(src_idx), src,
conf["sample_rate"])
for src_idx, est_src in enumerate(est_sources_np):
est_src *= np.max(np.abs(mix_np)) / np.max(np.abs(est_src))
sf.write(
local_save_dir + "s{}_estimate.wav".format(src_idx),
est_src,
conf["sample_rate"],
)
# Write local metrics to the example folder.
with open(local_save_dir + "metrics.json", "w") as f:
json.dump(utt_metrics, f, indent=0)
# Save all metrics to the experiment folder.
all_metrics_df = pd.DataFrame(series_list)
all_metrics_df.to_csv(os.path.join(eval_save_dir, "all_metrics.csv"))
# Print and save summary metrics
final_results = {}
for metric_name in compute_metrics:
input_metric_name = "input_" + metric_name
ldf = all_metrics_df[metric_name] - all_metrics_df[input_metric_name]
final_results[metric_name] = all_metrics_df[metric_name].mean()
final_results[metric_name + "_imp"] = ldf.mean()
print("Overall metrics :")
pprint(final_results)
with open(os.path.join(eval_save_dir, "final_metrics.json"), "w") as f:
json.dump(final_results, f, indent=0)
model_dict = torch.load(model_path, map_location="cpu")
os.makedirs(os.path.join(conf["exp_dir"], "publish_dir"), exist_ok=True)
# publishable = save_publishable(
save_publishable(
os.path.join(conf["exp_dir"], "publish_dir"),
model_dict,
metrics=final_results,
train_conf=train_conf,
)
# Handle device placement
if conf["use_gpu"]:
model.cuda()
model_device = next(model.parameters()).device
<<<<<<< HEAD
test_set = WhamDataset(
conf["test_dir"],
conf["task"],
sample_rate=conf["sample_rate"],
nondefault_nsrc=model.masker.n_src,
segment=None,
=======
test_set = LibriMix(
csv_dir=conf["test_dir"],
task=conf["task"],
sample_rate=conf["sample_rate"],
n_src=conf["train_conf"]["data"]["n_src"],
segment=None,
return_id=True,
>>>>>>> 210b5e4eb8ce24fe25780e008c89a4bb71bbd0ea
) # Uses all segment length
# Used to reorder sources only
loss_func = PITLossWrapper(pairwise_neg_sisdr, pit_from="pw_mtx")
# Randomly choose the indexes of sentences to save.
<<<<<<< HEAD
ex_save_dir = os.path.join(conf["exp_dir"], "examples/")
=======
eval_save_dir = os.path.join(conf["exp_dir"], conf["out_dir"])
ex_save_dir = os.path.join(eval_save_dir, "examples/")
>>>>>>> 210b5e4eb8ce24fe25780e008c89a4bb71bbd0ea