From 1988fe85f6d4e2d2f809e705e13d69d0b57bd0fc Mon Sep 17 00:00:00 2001
From: shixian.shi <shixian.shi@alibaba-inc.com>
Date: 星期四, 04 五月 2023 19:27:00 +0800
Subject: [PATCH] update
---
funasr/models/e2e_asr_contextual_paraformer.py | 20 +++-----------------
1 files changed, 3 insertions(+), 17 deletions(-)
diff --git a/funasr/models/e2e_asr_contextual_paraformer.py b/funasr/models/e2e_asr_contextual_paraformer.py
index cafb653..93027ec 100644
--- a/funasr/models/e2e_asr_contextual_paraformer.py
+++ b/funasr/models/e2e_asr_contextual_paraformer.py
@@ -1,4 +1,3 @@
-from json import decoder
import logging
from contextlib import contextmanager
from distutils.version import LooseVersion
@@ -7,35 +6,24 @@
from typing import Optional
from typing import Tuple
from typing import Union
-import random
-from unicodedata import bidirectional
import numpy as np
import torch
from typeguard import check_argument_types
from funasr.layers.abs_normalize import AbsNormalize
-from funasr.losses.label_smoothing_loss import (
- LabelSmoothingLoss, # noqa: H301
-)
from funasr.models.ctc import CTC
from funasr.models.decoder.abs_decoder import AbsDecoder
-from funasr.models.e2e_asr_common import ErrorCalculator
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.predictor.cif import mae_loss
from funasr.models.preencoder.abs_preencoder import AbsPreEncoder
from funasr.models.specaug.abs_specaug import AbsSpecAug
from funasr.modules.add_sos_eos import add_sos_eos
from funasr.modules.nets_utils import make_pad_mask, pad_list
from funasr.modules.nets_utils import th_accuracy
from funasr.torch_utils.device_funcs import force_gatherable
-from funasr.train.abs_espnet_model import AbsESPnetModel
-from funasr.models.predictor.cif import CifPredictorV3
-from funasr.modules.streaming_utils import utils as myutils
from funasr.models.e2e_asr_paraformer import Paraformer
-from funasr.modules.layer_norm import LayerNorm
if LooseVersion(torch.__version__) >= LooseVersion("1.6.0"):
@@ -47,7 +35,7 @@
yield
-class AdvancedContextualParaformer(Paraformer):
+class NeatContextualParaformer(Paraformer):
def __init__(
self,
vocab_size: int,
@@ -80,7 +68,7 @@
target_buffer_length: int = -1,
inner_dim: int = 256,
bias_encoder_type: str = 'lstm',
- use_decoder_embedding: bool = True,
+ use_decoder_embedding: bool = False,
crit_attn_weight: float = 0.0,
crit_attn_smooth: float = 0.0,
bias_encoder_dropout_rate: float = 0.0,
@@ -352,7 +340,7 @@
input_mask_expand_dim, 0)
return sematic_embeds * tgt_mask, decoder_out * tgt_mask
- def cal_decoder_with_predictor_with_hwlist_advanced(self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, hw_list=None):
+ def cal_decoder_with_predictor(self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, hw_list=None):
if hw_list is None:
hw_list = [torch.Tensor([1]).long().to(encoder_out.device)] # empty hotword list
hw_list_pad = pad_list(hw_list, 0)
@@ -362,7 +350,6 @@
hw_embed = self.bias_embed(hw_list_pad)
hw_embed, (h_n, _) = self.bias_encoder(hw_embed)
else:
- # hw_list = hw_list[1:] + [hw_list[0]] # reorder
hw_lengths = [len(i) for i in hw_list]
hw_list_pad = pad_list([torch.Tensor(i).long() for i in hw_list], 0).to(encoder_out.device)
if self.use_decoder_embedding:
@@ -378,7 +365,6 @@
if _h_n is not None:
h_n = _h_n
hw_embed = h_n.repeat(encoder_out.shape[0], 1, 1)
- # import pdb; pdb.set_trace()
decoder_outs = self.decoder(
encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, contextual_info=hw_embed
--
Gitblit v1.9.1