From c9f1b4e8a2e903f74de20d019e70307c26e93c3e Mon Sep 17 00:00:00 2001
From: shixian.shi <shixian.shi@alibaba-inc.com>
Date: 星期四, 23 十一月 2023 20:39:52 +0800
Subject: [PATCH] update

---
 funasr/models/e2e_asr_contextual_paraformer.py |    7 ++++---
 1 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/funasr/models/e2e_asr_contextual_paraformer.py b/funasr/models/e2e_asr_contextual_paraformer.py
index 4836663..171a6c6 100644
--- a/funasr/models/e2e_asr_contextual_paraformer.py
+++ b/funasr/models/e2e_asr_contextual_paraformer.py
@@ -134,7 +134,7 @@
             text_lengths: torch.Tensor,
             hotword_pad: torch.Tensor,
             hotword_lengths: torch.Tensor,
-            ideal_attn: torch.Tensor,
+            dha_pad: torch.Tensor,
     ) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
         """Frontend + Encoder + Decoder + Calc loss
 
@@ -341,7 +341,7 @@
             input_mask_expand_dim, 0)
         return sematic_embeds * tgt_mask, decoder_out * tgt_mask
 
-    def cal_decoder_with_predictor(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, clas_scale=1.0):
         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)
@@ -350,6 +350,7 @@
             else:
                 hw_embed = self.bias_embed(hw_list_pad)
             hw_embed, (h_n, _) = self.bias_encoder(hw_embed)
+            hw_embed = h_n.repeat(encoder_out.shape[0], 1, 1)
         else:
             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)
@@ -363,7 +364,7 @@
             hw_embed = h_n.repeat(encoder_out.shape[0], 1, 1)
         
         decoder_outs = self.decoder(
-            encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, contextual_info=hw_embed
+            encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, contextual_info=hw_embed, clas_scale=clas_scale
         )
         decoder_out = decoder_outs[0]
         decoder_out = torch.log_softmax(decoder_out, dim=-1)

--
Gitblit v1.9.1