|
|
|
"""*********************************************************************************************""" |
|
|
|
|
|
|
|
|
|
"""*********************************************************************************************""" |
|
|
|
|
|
|
|
|
|
|
|
import os |
|
import math |
|
import torch |
|
import random |
|
|
|
import torch |
|
import torch.nn as nn |
|
from torch.utils.data import DataLoader |
|
from torch.nn.utils.rnn import pad_sequence |
|
|
|
from argparse import Namespace |
|
from ..voxceleb1.expert import DownstreamExpert as SpeakerExpert |
|
|
|
class DownstreamExpert(SpeakerExpert): |
|
""" |
|
Used to handle downstream-specific operations |
|
eg. downstream forward, metric computation, contents to log |
|
""" |
|
|
|
def __init__(self, upstream_dim, downstream_expert, expdir, **kwargs): |
|
super(DownstreamExpert, self).__init__(upstream_dim, downstream_expert, expdir, **kwargs) |
|
|
|
def forward(self, mode, features, lengths, labels, records, **kwargs): |
|
device = features[0].device |
|
features_len = torch.IntTensor([len(feat) for feat in features]).to(device=device) |
|
features = pad_sequence(features, batch_first=True) |
|
features = self.projector(features) |
|
predicted, _ = self.model(features, features_len) |
|
|
|
labels = torch.LongTensor(labels).to(features.device) |
|
|
|
predicted = predicted.transpose(-1,-2) |
|
labels = [labels[index].expand(features_len[index]) for index in range(len(labels))] |
|
labels = pad_sequence(labels, padding_value=-100, batch_first=True) |
|
|
|
loss = self.objective(predicted, labels) |
|
|
|
predicted = predicted.transpose(-1,-2) |
|
predicted_classid = predicted.max(dim=-1).indices |
|
records['acc'] += (predicted_classid[labels!=-100] == labels[labels!=-100]).view(-1).cpu().float().tolist() |
|
records['loss'].append(loss.item()) |
|
|
|
return loss |
|
|
