|
from speaker_encoder.params_model import * |
|
from speaker_encoder.params_data import * |
|
from scipy.interpolate import interp1d |
|
from sklearn.metrics import roc_curve |
|
from torch.nn.utils import clip_grad_norm_ |
|
from scipy.optimize import brentq |
|
from torch import nn |
|
import numpy as np |
|
import torch |
|
|
|
|
|
class SpeakerEncoder(nn.Module): |
|
def __init__(self, device, loss_device): |
|
super().__init__() |
|
self.loss_device = loss_device |
|
|
|
|
|
self.lstm = nn.LSTM(input_size=mel_n_channels, |
|
hidden_size=model_hidden_size, |
|
num_layers=model_num_layers, |
|
batch_first=True).to(device) |
|
self.linear = nn.Linear(in_features=model_hidden_size, |
|
out_features=model_embedding_size).to(device) |
|
self.relu = torch.nn.ReLU().to(device) |
|
|
|
|
|
self.similarity_weight = nn.Parameter(torch.tensor([10.])).to(loss_device) |
|
self.similarity_bias = nn.Parameter(torch.tensor([-5.])).to(loss_device) |
|
|
|
|
|
self.loss_fn = nn.CrossEntropyLoss().to(loss_device) |
|
|
|
def do_gradient_ops(self): |
|
|
|
self.similarity_weight.grad *= 0.01 |
|
self.similarity_bias.grad *= 0.01 |
|
|
|
|
|
clip_grad_norm_(self.parameters(), 3, norm_type=2) |
|
|
|
def forward(self, utterances, hidden_init=None): |
|
""" |
|
Computes the embeddings of a batch of utterance spectrograms. |
|
|
|
:param utterances: batch of mel-scale filterbanks of same duration as a tensor of shape |
|
(batch_size, n_frames, n_channels) |
|
:param hidden_init: initial hidden state of the LSTM as a tensor of shape (num_layers, |
|
batch_size, hidden_size). Will default to a tensor of zeros if None. |
|
:return: the embeddings as a tensor of shape (batch_size, embedding_size) |
|
""" |
|
|
|
|
|
out, (hidden, cell) = self.lstm(utterances, hidden_init) |
|
|
|
|
|
embeds_raw = self.relu(self.linear(hidden[-1])) |
|
|
|
|
|
embeds = embeds_raw / torch.norm(embeds_raw, dim=1, keepdim=True) |
|
|
|
return embeds |
|
|
|
def similarity_matrix(self, embeds): |
|
""" |
|
Computes the similarity matrix according the section 2.1 of GE2E. |
|
|
|
:param embeds: the embeddings as a tensor of shape (speakers_per_batch, |
|
utterances_per_speaker, embedding_size) |
|
:return: the similarity matrix as a tensor of shape (speakers_per_batch, |
|
utterances_per_speaker, speakers_per_batch) |
|
""" |
|
speakers_per_batch, utterances_per_speaker = embeds.shape[:2] |
|
|
|
|
|
centroids_incl = torch.mean(embeds, dim=1, keepdim=True) |
|
centroids_incl = centroids_incl.clone() / torch.norm(centroids_incl, dim=2, keepdim=True) |
|
|
|
|
|
centroids_excl = (torch.sum(embeds, dim=1, keepdim=True) - embeds) |
|
centroids_excl /= (utterances_per_speaker - 1) |
|
centroids_excl = centroids_excl.clone() / torch.norm(centroids_excl, dim=2, keepdim=True) |
|
|
|
|
|
|
|
|
|
sim_matrix = torch.zeros(speakers_per_batch, utterances_per_speaker, |
|
speakers_per_batch).to(self.loss_device) |
|
mask_matrix = 1 - np.eye(speakers_per_batch, dtype=np.int) |
|
for j in range(speakers_per_batch): |
|
mask = np.where(mask_matrix[j])[0] |
|
sim_matrix[mask, :, j] = (embeds[mask] * centroids_incl[j]).sum(dim=2) |
|
sim_matrix[j, :, j] = (embeds[j] * centroids_excl[j]).sum(dim=1) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sim_matrix = sim_matrix * self.similarity_weight + self.similarity_bias |
|
return sim_matrix |
|
|
|
def loss(self, embeds): |
|
""" |
|
Computes the softmax loss according the section 2.1 of GE2E. |
|
|
|
:param embeds: the embeddings as a tensor of shape (speakers_per_batch, |
|
utterances_per_speaker, embedding_size) |
|
:return: the loss and the EER for this batch of embeddings. |
|
""" |
|
speakers_per_batch, utterances_per_speaker = embeds.shape[:2] |
|
|
|
|
|
sim_matrix = self.similarity_matrix(embeds) |
|
sim_matrix = sim_matrix.reshape((speakers_per_batch * utterances_per_speaker, |
|
speakers_per_batch)) |
|
ground_truth = np.repeat(np.arange(speakers_per_batch), utterances_per_speaker) |
|
target = torch.from_numpy(ground_truth).long().to(self.loss_device) |
|
loss = self.loss_fn(sim_matrix, target) |
|
|
|
|
|
with torch.no_grad(): |
|
inv_argmax = lambda i: np.eye(1, speakers_per_batch, i, dtype=np.int)[0] |
|
labels = np.array([inv_argmax(i) for i in ground_truth]) |
|
preds = sim_matrix.detach().cpu().numpy() |
|
|
|
|
|
fpr, tpr, thresholds = roc_curve(labels.flatten(), preds.flatten()) |
|
eer = brentq(lambda x: 1. - x - interp1d(fpr, tpr)(x), 0., 1.) |
|
|
|
return loss, eer |