From 63eb03a28f88e70baf9c2a31c6991ae31a27640d Mon Sep 17 00:00:00 2001
From: lingyunfly <121302812+lingyunfly@users.noreply.github.com>
Date: 星期四, 09 二月 2023 15:02:04 +0800
Subject: [PATCH] Create vad_inference_launch.py
---
funasr/models/e2e_asr_paraformer.py | 6 +++---
1 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/funasr/models/e2e_asr_paraformer.py b/funasr/models/e2e_asr_paraformer.py
index 65c70df..7596896 100644
--- a/funasr/models/e2e_asr_paraformer.py
+++ b/funasr/models/e2e_asr_paraformer.py
@@ -499,11 +499,11 @@
def sampler(self, encoder_out, encoder_out_lens, ys_pad, ys_pad_lens, pre_acoustic_embeds):
tgt_mask = (~make_pad_mask(ys_pad_lens, maxlen=ys_pad_lens.max())[:, :, None]).to(ys_pad.device)
- ys_pad = ys_pad * tgt_mask[:, :, 0]
+ ys_pad_masked = ys_pad * tgt_mask[:, :, 0]
if self.share_embedding:
- ys_pad_embed = self.decoder.output_layer.weight[ys_pad]
+ ys_pad_embed = self.decoder.output_layer.weight[ys_pad_masked]
else:
- ys_pad_embed = self.decoder.embed(ys_pad)
+ ys_pad_embed = self.decoder.embed(ys_pad_masked)
with torch.no_grad():
decoder_outs = self.decoder(
encoder_out, encoder_out_lens, pre_acoustic_embeds, ys_pad_lens
--
Gitblit v1.9.1