From eb1574b813e230b156fc09eaaf03227b1b0b4134 Mon Sep 17 00:00:00 2001
From: weilikai <jasper@talkus.fun>
Date: 星期六, 20 九月 2025 22:41:05 +0800
Subject: [PATCH] fix: support loading .pcm (16k 1c 16bit) audio files in load_utils.py (#2667) (#2668)
---
funasr/models/sond/encoder/ci_scorers.py | 18 ++++++------------
1 files changed, 6 insertions(+), 12 deletions(-)
diff --git a/funasr/models/sond/encoder/ci_scorers.py b/funasr/models/sond/encoder/ci_scorers.py
index 50056ee..a60560b 100644
--- a/funasr/models/sond/encoder/ci_scorers.py
+++ b/funasr/models/sond/encoder/ci_scorers.py
@@ -7,17 +7,14 @@
super().__init__()
def forward(
- self,
- xs_pad: torch.Tensor,
- spk_emb: torch.Tensor,
+ self,
+ xs_pad: torch.Tensor,
+ spk_emb: torch.Tensor,
):
# xs_pad: B, T, D
# spk_emb: B, N, D
scores = torch.matmul(xs_pad, spk_emb.transpose(1, 2))
return scores
-
- def convert_tf2torch(self, var_dict_tf, var_dict_torch):
- return {}
class CosScorer(torch.nn.Module):
@@ -25,14 +22,11 @@
super().__init__()
def forward(
- self,
- xs_pad: torch.Tensor,
- spk_emb: torch.Tensor,
+ self,
+ xs_pad: torch.Tensor,
+ spk_emb: torch.Tensor,
):
# xs_pad: B, T, D
# spk_emb: B, N, D
scores = F.cosine_similarity(xs_pad.unsqueeze(2), spk_emb.unsqueeze(1), dim=-1)
return scores
-
- def convert_tf2torch(self, var_dict_tf, var_dict_torch):
- return {}
--
Gitblit v1.9.1