update

speech_asr

2023-04-20 3e77fd44304a67a2b2253b4e56fede9762bb8464

 funasr/models/e2e_uni_asr.py

@@ -18,14 +18,11 @@
)
from funasr.models.ctc import CTC
from funasr.models.decoder.abs_decoder import AbsDecoder
from funasr.models.encoder.abs_encoder import AbsEncoder
from funasr.models.frontend.abs_frontend import AbsFrontend
from funasr.models.postencoder.abs_postencoder import AbsPostEncoder
from funasr.models.preencoder.abs_preencoder import AbsPreEncoder
from funasr.models.specaug.abs_specaug import AbsSpecAug
from funasr.layers.abs_normalize import AbsNormalize
from funasr.models.base_model import FunASRModel
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.modules.streaming_utils.chunk_utilis import sequence_mask
from funasr.models.predictor.cif import mae_loss

@@ -38,7 +35,7 @@
        yield


class UniASR(AbsESPnetModel):
class UniASR(FunASRModel):
    """
    Author: Speech Lab, Alibaba Group, China
    """
@@ -47,11 +44,11 @@
        self,
        vocab_size: int,
        token_list: Union[Tuple[str, ...], List[str]],
        frontend: Optional[AbsFrontend],
        specaug: Optional[AbsSpecAug],
        normalize: Optional[AbsNormalize],
        frontend: Optional[torch.nn.Module],
        specaug: Optional[torch.nn.Module],
        normalize: Optional[torch.nn.Module],
        preencoder: Optional[AbsPreEncoder],
        encoder: AbsEncoder,
        encoder: torch.nn.Module,
        postencoder: Optional[AbsPostEncoder],
        decoder: AbsDecoder,
        ctc: CTC,
@@ -206,7 +203,7 @@
            with torch.no_grad():
                speech_raw, encoder_out, encoder_out_lens = self.encode(speech, speech_lengths, ind=ind)
        else:
            speech_raw, encoder_out_lens = self.encode(speech, speech_lengths, ind=ind)
            speech_raw, encoder_out, encoder_out_lens = self.encode(speech, speech_lengths, ind=ind)

        intermediate_outs = None
        if isinstance(encoder_out, tuple):