From 54931dd4e1a099d7d6f144c4e12e5453deb3aa26 Mon Sep 17 00:00:00 2001
From: 雾聪 <wucong.lyb@alibaba-inc.com>
Date: 星期三, 28 六月 2023 10:41:57 +0800
Subject: [PATCH] Merge branch 'main' of https://github.com/alibaba-damo-academy/FunASR into main
---
funasr/models/e2e_diar_sond.py | 118 +++++++++++++++++++++++++++++++++++-----------------------
1 files changed, 71 insertions(+), 47 deletions(-)
diff --git a/funasr/models/e2e_diar_sond.py b/funasr/models/e2e_diar_sond.py
index 7b6e955..9c3fb92 100644
--- a/funasr/models/e2e_diar_sond.py
+++ b/funasr/models/e2e_diar_sond.py
@@ -22,7 +22,7 @@
from funasr.models.specaug.abs_specaug import AbsSpecAug
from funasr.layers.abs_normalize import AbsNormalize
from funasr.torch_utils.device_funcs import force_gatherable
-from funasr.train.abs_espnet_model import AbsESPnetModel
+from funasr.models.base_model import FunASRModel
from funasr.losses.label_smoothing_loss import LabelSmoothingLoss, SequenceBinaryCrossEntropy
from funasr.utils.misc import int2vec
@@ -35,9 +35,13 @@
yield
-class DiarSondModel(AbsESPnetModel):
- """Speaker overlap-aware neural diarization model
- reference: https://arxiv.org/abs/2211.10243
+class DiarSondModel(FunASRModel):
+ """
+ Author: Speech Lab, Alibaba Group, China
+ SOND: Speaker Overlap-aware Neural Diarization for Multi-party Meeting Analysis
+ https://arxiv.org/abs/2211.10243
+ TOLD: A Novel Two-Stage Overlap-Aware Framework for Speaker Diarization
+ https://arxiv.org/abs/2303.05397
"""
def __init__(
@@ -46,10 +50,10 @@
frontend: Optional[AbsFrontend],
specaug: Optional[AbsSpecAug],
normalize: Optional[AbsNormalize],
- encoder: AbsEncoder,
- speaker_encoder: AbsEncoder,
+ encoder: torch.nn.Module,
+ speaker_encoder: Optional[torch.nn.Module],
ci_scorer: torch.nn.Module,
- cd_scorer: torch.nn.Module,
+ cd_scorer: Optional[torch.nn.Module],
decoder: torch.nn.Module,
token_list: list,
lsm_weight: float = 0.1,
@@ -59,7 +63,8 @@
normalize_speech_speaker: bool = False,
ignore_id: int = -1,
speaker_discrimination_loss_weight: float = 1.0,
- inter_score_loss_weight: float = 0.0
+ inter_score_loss_weight: float = 0.0,
+ inputs_type: str = "raw",
):
assert check_argument_types()
@@ -86,13 +91,17 @@
)
self.criterion_bce = SequenceBinaryCrossEntropy(normalize_length=length_normalized_loss)
self.pse_embedding = self.generate_pse_embedding()
+ self.power_weight = torch.from_numpy(2 ** np.arange(max_spk_num)[np.newaxis, np.newaxis, :]).float()
+ self.int_token_arr = torch.from_numpy(np.array(self.token_list).astype(int)[np.newaxis, np.newaxis, :]).int()
self.speaker_discrimination_loss_weight = speaker_discrimination_loss_weight
self.inter_score_loss_weight = inter_score_loss_weight
+ self.forward_steps = 0
+ self.inputs_type = inputs_type
def generate_pse_embedding(self):
embedding = np.zeros((len(self.token_list), self.max_spk_num), dtype=np.float)
for idx, pse_label in enumerate(self.token_list):
- emb = int2vec(pse_label, vec_dim=self.max_spk_num, dtype=np.float)
+ emb = int2vec(int(pse_label), vec_dim=self.max_spk_num, dtype=np.float)
embedding[idx] = emb
return torch.from_numpy(embedding)
@@ -102,11 +111,10 @@
speech_lengths: torch.Tensor = None,
profile: torch.Tensor = None,
profile_lengths: torch.Tensor = None,
- spk_labels: torch.Tensor = None,
- spk_labels_lengths: torch.Tensor = None,
+ binary_labels: torch.Tensor = None,
+ binary_labels_lengths: torch.Tensor = None,
) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
"""Frontend + Encoder + Speaker Encoder + CI Scorer + CD Scorer + Decoder + Calc loss
-
Args:
speech: (Batch, samples) or (Batch, frames, input_size)
speech_lengths: (Batch,) default None for chunk interator,
@@ -116,11 +124,16 @@
espnet2/iterators/chunk_iter_factory.py
profile: (Batch, N_spk, dim)
profile_lengths: (Batch,)
- spk_labels: (Batch, frames, input_size)
- spk_labels_lengths: (Batch,)
+ binary_labels: (Batch, frames, max_spk_num)
+ binary_labels_lengths: (Batch,)
"""
- assert speech.shape[0] == spk_labels.shape[0], (speech.shape, spk_labels.shape)
+ assert speech.shape[0] <= binary_labels.shape[0], (speech.shape, binary_labels.shape)
batch_size = speech.shape[0]
+ self.forward_steps = self.forward_steps + 1
+ if self.pse_embedding.device != speech.device:
+ self.pse_embedding = self.pse_embedding.to(speech.device)
+ self.power_weight = self.power_weight.to(speech.device)
+ self.int_token_arr = self.int_token_arr.to(speech.device)
# 1. Network forward
pred, inter_outputs = self.prediction_forward(
@@ -132,23 +145,29 @@
# 2. Aggregate time-domain labels to match forward outputs
if self.label_aggregator is not None:
- spk_labels, spk_labels_lengths = self.label_aggregator(
- spk_labels.unsqueeze(2), spk_labels_lengths
+ binary_labels, binary_labels_lengths = self.label_aggregator(
+ binary_labels, binary_labels_lengths
)
- spk_labels = spk_labels.squeeze(2)
+ # 2. Calculate power-set encoding (PSE) labels
+ raw_pse_labels = torch.sum(binary_labels * self.power_weight, dim=2, keepdim=True)
+ pse_labels = torch.argmax((raw_pse_labels.int() == self.int_token_arr).float(), dim=2)
# If encoder uses conv* as input_layer (i.e., subsampling),
# the sequence length of 'pred' might be slightly less than the
# length of 'spk_labels'. Here we force them to be equal.
length_diff_tolerance = 2
- length_diff = spk_labels.shape[1] - pred.shape[1]
- if 0 < length_diff <= length_diff_tolerance:
- spk_labels = spk_labels[:, 0: pred.shape[1], :]
+ length_diff = abs(pse_labels.shape[1] - pred.shape[1])
+ if length_diff <= length_diff_tolerance:
+ min_len = min(pred.shape[1], pse_labels.shape[1])
+ pse_labels = pse_labels[:, :min_len]
+ pred = pred[:, :min_len]
+ cd_score = cd_score[:, :min_len]
+ ci_score = ci_score[:, :min_len]
- loss_diar = self.classification_loss(pred, spk_labels, spk_labels_lengths)
+ loss_diar = self.classification_loss(pred, pse_labels, binary_labels_lengths)
loss_spk_dis = self.speaker_discrimination_loss(profile, profile_lengths)
- loss_inter_ci, loss_inter_cd = self.internal_score_loss(cd_score, ci_score, spk_labels, spk_labels_lengths)
- label_mask = make_pad_mask(spk_labels_lengths, maxlen=spk_labels.shape[1])
+ loss_inter_ci, loss_inter_cd = self.internal_score_loss(cd_score, ci_score, pse_labels, binary_labels_lengths)
+ label_mask = make_pad_mask(binary_labels_lengths, maxlen=pse_labels.shape[1]).to(pse_labels.device)
loss = (loss_diar + self.speaker_discrimination_loss_weight * loss_spk_dis
+ self.inter_score_loss_weight * (loss_inter_ci + loss_inter_cd))
@@ -163,9 +182,9 @@
speaker_falarm,
speaker_error,
) = self.calc_diarization_error(
- pred=F.embedding(pred.argmax(dim=2) * label_mask, self.pse_embedding),
- label=F.embedding(spk_labels * label_mask, self.pse_embedding),
- length=spk_labels_lengths
+ pred=F.embedding(pred.argmax(dim=2) * (~label_mask), self.pse_embedding),
+ label=F.embedding(pse_labels * (~label_mask), self.pse_embedding),
+ length=binary_labels_lengths
)
if speech_scored > 0 and num_frames > 0:
@@ -194,6 +213,7 @@
cf=cf,
acc=acc,
der=der,
+ forward_steps=self.forward_steps,
)
loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)
@@ -205,11 +225,12 @@
labels: torch.Tensor,
prediction_lengths: torch.Tensor
) -> torch.Tensor:
+ mask = make_pad_mask(prediction_lengths, maxlen=labels.shape[1])
pad_labels = labels.masked_fill(
- make_pad_mask(prediction_lengths, maxlen=labels.shape[1]),
+ mask.to(predictions.device),
value=self.ignore_id
)
- loss = self.criterion_diar(predictions, pad_labels)
+ loss = self.criterion_diar(predictions.contiguous(), pad_labels)
return loss
@@ -220,24 +241,26 @@
) -> torch.Tensor:
profile_mask = (torch.linalg.norm(profile, ord=2, dim=2, keepdim=True) > 0).float() # (B, N, 1)
mask = torch.matmul(profile_mask, profile_mask.transpose(1, 2)) # (B, N, N)
- mask = mask * (1.0 - torch.eye(self.max_spk_num).unsqueeze(0))
+ mask = mask * (1.0 - torch.eye(self.max_spk_num).unsqueeze(0).to(mask))
eps = 1e-12
coding_norm = torch.linalg.norm(
profile * profile_mask + (1 - profile_mask) * eps,
dim=2, keepdim=True
) * profile_mask
- cos_theta = F.cosine_similarity(profile, profile, dim=2, eps=eps) * mask
+ # profile: Batch, N, dim
+ cos_theta = F.cosine_similarity(profile.unsqueeze(2), profile.unsqueeze(1), dim=-1, eps=eps) * mask
cos_theta = torch.clip(cos_theta, -1 + eps, 1 - eps)
loss = (F.relu(mask * coding_norm * (cos_theta - 0.0))).sum() / mask.sum()
return loss
def calculate_multi_labels(self, pse_labels, pse_labels_lengths):
+ mask = make_pad_mask(pse_labels_lengths, maxlen=pse_labels.shape[1])
padding_labels = pse_labels.masked_fill(
- make_pad_mask(pse_labels_lengths, maxlen=pse_labels.shape[1]),
+ mask.to(pse_labels.device),
value=0
- ).to(pse_labels.dtype)
+ ).to(pse_labels)
multi_labels = F.embedding(padding_labels, self.pse_embedding)
return multi_labels
@@ -258,8 +281,10 @@
self,
speech: torch.Tensor,
speech_lengths: torch.Tensor,
- spk_labels: torch.Tensor = None,
- spk_labels_lengths: torch.Tensor = None,
+ profile: torch.Tensor = None,
+ profile_lengths: torch.Tensor = None,
+ binary_labels: torch.Tensor = None,
+ binary_labels_lengths: torch.Tensor = None,
) -> Dict[str, torch.Tensor]:
feats, feats_lengths = self._extract_feats(speech, speech_lengths)
return {"feats": feats, "feats_lengths": feats_lengths}
@@ -285,7 +310,7 @@
speech: torch.Tensor,
speech_lengths: torch.Tensor,
) -> Tuple[torch.Tensor, torch.Tensor]:
- if self.encoder is not None:
+ if self.encoder is not None and self.inputs_type == "raw":
speech, speech_lengths = self.encode(speech, speech_lengths)
speech_mask = ~make_pad_mask(speech_lengths, maxlen=speech.shape[1])
speech_mask = speech_mask.to(speech.device).unsqueeze(-1).float()
@@ -312,7 +337,7 @@
speaker_encoder_outputs: torch.Tensor,
seq_len: torch.Tensor = None,
spk_len: torch.Tensor = None,
- ) -> torch.Tensor:
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
bb, tt = speech_encoder_outputs.shape[0], speech_encoder_outputs.shape[1]
d_sph, d_spk = speech_encoder_outputs.shape[2], speaker_encoder_outputs.shape[2]
if self.normalize_speech_speaker:
@@ -328,11 +353,11 @@
if isinstance(self.ci_scorer, AbsEncoder):
ci_simi = self.ci_scorer(ge_in, ge_len)[0]
+ ci_simi = torch.reshape(ci_simi, [bb, self.max_spk_num, tt]).permute([0, 2, 1])
else:
ci_simi = self.ci_scorer(speech_encoder_outputs, speaker_encoder_outputs)
- simi = torch.cat([cd_simi, ci_simi], dim=2)
- return simi
+ return ci_simi, cd_simi
def post_net_forward(self, simi, seq_len):
logits = self.decoder(simi, seq_len)[0]
@@ -352,19 +377,19 @@
# speaker encoding
profile, profile_lengths = self.encode_speaker(profile, profile_lengths)
# calculating similarity
- similarity = self.calc_similarity(speech, profile, speech_lengths, profile_lengths)
+ ci_simi, cd_simi = self.calc_similarity(speech, profile, speech_lengths, profile_lengths)
+ similarity = torch.cat([cd_simi, ci_simi], dim=2)
# post net forward
logits = self.post_net_forward(similarity, speech_lengths)
if return_inter_outputs:
- return logits, [(speech, speech_lengths), (profile, profile_lengths), torch.split(similarity, 2)]
+ return logits, [(speech, speech_lengths), (profile, profile_lengths), (ci_simi, cd_simi)]
return logits
def encode(
self, speech: torch.Tensor, speech_lengths: torch.Tensor
) -> Tuple[torch.Tensor, torch.Tensor]:
"""Frontend + Encoder
-
Args:
speech: (Batch, Length, ...)
speech_lengths: (Batch,)
@@ -384,7 +409,8 @@
# 4. Forward encoder
# feats: (Batch, Length, Dim)
# -> encoder_out: (Batch, Length2, Dim)
- encoder_out, encoder_out_lens, _ = self.encoder(feats, feats_lengths)
+ encoder_outputs = self.encoder(feats, feats_lengths)
+ encoder_out, encoder_out_lens = encoder_outputs[:2]
assert encoder_out.size(0) == speech.size(0), (
encoder_out.size(),
@@ -429,9 +455,7 @@
(batch_size, max_len, num_output) = label.size()
# mask the padding part
- mask = np.zeros((batch_size, max_len, num_output))
- for i in range(batch_size):
- mask[i, : length[i], :] = 1
+ mask = ~make_pad_mask(length, maxlen=label.shape[1]).unsqueeze(-1).numpy()
# pred and label have the shape (batch_size, max_len, num_output)
label_np = label.data.cpu().numpy().astype(int)
@@ -465,4 +489,4 @@
speaker_miss,
speaker_falarm,
speaker_error,
- )
+ )
\ No newline at end of file
--
Gitblit v1.9.1