update eend_ola

egs/callhome/eend_ola/conf/train_diar_eend_ola_callhome_chunk2000.yaml

@@ -12,7 +12,7 @@ encoder_decoder_attractor_conf:
     n_units: 256
 
 # model related
-model: eend_ola_similar_eend
+model: eend_ola
 model_conf:
     attractor_loss_weight:  0.01
     max_n_speaker: 8

egs/callhome/eend_ola/conf/train_diar_eend_ola_simu_2spkr.yaml

@@ -12,7 +12,7 @@ encoder_decoder_attractor_conf:
     n_units: 256
 
 # model related
-model: eend_ola_similar_eend
+model: eend_ola
 model_conf:
     max_n_speaker: 8
 

egs/callhome/eend_ola/conf/train_diar_eend_ola_simu_allspkr.yaml

@@ -12,7 +12,7 @@ encoder_decoder_attractor_conf:
     n_units: 256
 
 # model related
-model: eend_ola_similar_eend
+model: eend_ola
 model_conf:
     max_n_speaker: 8
 

egs/callhome/eend_ola/conf/train_diar_eend_ola_simu_allspkr_chunk2000.yaml

@@ -12,7 +12,7 @@ encoder_decoder_attractor_conf:
     n_units: 256
 
 # model related
-model: eend_ola_similar_eend
+model: eend_ola
 model_conf:
     max_n_speaker: 8
 

funasr/models/e2e_diar_eend_ola.py

@@ -12,7 +12,7 @@ from funasr.models.base_model import FunASRModel
 from funasr.models.frontend.wav_frontend import WavFrontendMel23
 from funasr.modules.eend_ola.encoder import EENDOLATransformerEncoder
 from funasr.modules.eend_ola.encoder_decoder_attractor import EncoderDecoderAttractor
-from funasr.modules.eend_ola.utils.losses import fast_batch_pit_n_speaker_loss, standard_loss, cal_power_loss
+from funasr.modules.eend_ola.utils.losses import standard_loss, cal_power_loss, fast_batch_pit_n_speaker_loss
 from funasr.modules.eend_ola.utils.power import create_powerlabel
 from funasr.modules.eend_ola.utils.power import generate_mapping_dict
 from funasr.torch_utils.device_funcs import force_gatherable
@@ -109,23 +109,17 @@ class DiarEENDOLAModel(FunASRModel):
     def forward(
             self,
             speech: List[torch.Tensor],
-            speech_lengths: torch.Tensor,  # num_frames of each sample
             speaker_labels: List[torch.Tensor],
-            speaker_labels_lengths: torch.Tensor,  # num_speakers of each sample
             orders: torch.Tensor,
     ) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
 
         # Check that batch_size is unified
-        assert (
-                len(speech)
-                == len(speech_lengths)
-                == len(speaker_labels)
-                == len(speaker_labels_lengths)
-        ), (len(speech), len(speech_lengths), len(speaker_labels), len(speaker_labels_lengths))
+        assert (len(speech) == len(speaker_labels)), (len(speech), len(speaker_labels))
+        speech_lengths = torch.tensor([len(sph) for sph in speech]).to(torch.int64)
+        speaker_labels_lengths = torch.tensor([spk.shape[-1] for spk in speaker_labels]).to(torch.int64)
         batch_size = len(speech)
 
         # Encoder
-        speech = [s[:s_len] for s, s_len in zip(speech, speech_lengths)]
         encoder_out = self.forward_encoder(speech, speech_lengths)
 
         # Encoder-decoder attractor