From 810046e3df4910c8f5c1a90e4b53aca45b3397e8 Mon Sep 17 00:00:00 2001
From: wuhongsheng <664116298@qq.com>
Date: 星期一, 01 七月 2024 10:42:58 +0800
Subject: [PATCH] 优化merge segments 参数，解决新闻联播男女主持人“晚上好”合并一个speakid问题 (#1861)
---
funasr/models/sense_voice/whisper_lib/decoding.py | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/funasr/models/sense_voice/whisper_lib/decoding.py b/funasr/models/sense_voice/whisper_lib/decoding.py
index 609d6a6..a468efa 100644
--- a/funasr/models/sense_voice/whisper_lib/decoding.py
+++ b/funasr/models/sense_voice/whisper_lib/decoding.py
@@ -63,8 +63,8 @@
else:
x = x.to(mel.device)
# FIX(funasr): sense vocie
- # logits = model.logits(x[:, :-1], mel)[:, -1]
- logits = model.logits(x[:, :], mel)[:, -1]
+ logits = model.logits(x[:, :-1], mel)[:, -1]
+ # logits = model.logits(x[:, :], mel)[:, -1]
# collect detected languages; suppress all non-language tokens
mask = torch.ones(logits.shape[-1], dtype=torch.bool)
--
Gitblit v1.9.1