update

speech_asr

2023-04-11 d5a80d642a5721eb1352cba59833a5cf4b91000f

 funasr/models/e2e_tp.py

@@ -14,10 +14,10 @@
from funasr.models.encoder.abs_encoder import AbsEncoder
from funasr.models.frontend.abs_frontend import AbsFrontend
from funasr.models.predictor.cif import mae_loss
from funasr.models.base_model import FunASRModel
from funasr.modules.add_sos_eos import add_sos_eos
from funasr.modules.nets_utils import make_pad_mask, pad_list
from funasr.torch_utils.device_funcs import force_gatherable
from funasr.train.abs_espnet_model import AbsESPnetModel
from funasr.models.predictor.cif import CifPredictorV3


@@ -30,7 +30,7 @@
        yield


class TimestampPredictor(AbsESPnetModel):
class TimestampPredictor(FunASRModel):
    """
    Author: Speech Lab, Alibaba Group, China
    """
@@ -150,10 +150,10 @@
    def calc_predictor_timestamp(self, encoder_out, encoder_out_lens, token_num):
        encoder_out_mask = (~make_pad_mask(encoder_out_lens, maxlen=encoder_out.size(1))[:, None, :]).to(
            encoder_out.device)
        ds_alphas, ds_cif_peak, us_alphas, us_cif_peak = self.predictor.get_upsample_timestamp(encoder_out,
        ds_alphas, ds_cif_peak, us_alphas, us_peaks = self.predictor.get_upsample_timestamp(encoder_out,
                                                                                               encoder_out_mask,
                                                                                               token_num)
        return ds_alphas, ds_cif_peak, us_alphas, us_cif_peak
        return ds_alphas, ds_cif_peak, us_alphas, us_peaks

    def collect_feats(
            self,