From a32a2b893436fad56cc82409121c7fa36a1815d5 Mon Sep 17 00:00:00 2001
From: RVC-Boss <129054828+RVC-Boss@users.noreply.github.com>
Date: Fri, 28 Feb 2025 16:50:35 +0800
Subject: [PATCH] =?UTF-8?q?v3sovits=E6=A8=A1=E5=9E=8B=E6=8E=A8=E7=90=86?=
 =?UTF-8?q?=E6=94=AF=E6=8C=81webui=E4=BC=A0=E8=AF=AD=E9=80=9F=E5=8F=82?=
 =?UTF-8?q?=E6=95=B0=E8=B0=83=E6=95=B4=E5=90=88=E6=88=90=E8=AF=AD=E9=80=9F?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

v3sovits模型推理支持webui传语速参数调整合成语速
---
 GPT_SoVITS/module/models.py | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/GPT_SoVITS/module/models.py b/GPT_SoVITS/module/models.py
index 338e88d..623da80 100644
--- a/GPT_SoVITS/module/models.py
+++ b/GPT_SoVITS/module/models.py
@@ -1251,7 +1251,7 @@ class SynthesizerTrnV3(nn.Module):
         return cfm_loss
 
     @torch.no_grad()
-    def decode_encp(self, codes,text, refer,ge=None):
+    def decode_encp(self, codes,text, refer,ge=None,speed=1):
         # print(2333333,refer.shape)
         # ge=None
         if(ge==None):
@@ -1259,13 +1259,17 @@ class SynthesizerTrnV3(nn.Module):
             refer_mask = torch.unsqueeze(commons.sequence_mask(refer_lengths, refer.size(2)), 1).to(refer.dtype)
             ge = self.ref_enc(refer[:,:704] * refer_mask, refer_mask)
         y_lengths = torch.LongTensor([int(codes.size(2)*2)]).to(codes.device)
-        y_lengths1 = torch.LongTensor([int(codes.size(2)*2.5*1.5)]).to(codes.device)
+        if speed==1:
+            sizee=int(codes.size(2)*2.5*1.5)
+        else:
+            sizee=int(codes.size(2)*2.5*1.5/speed)+1
+        y_lengths1 = torch.LongTensor([sizee]).to(codes.device)
         text_lengths = torch.LongTensor([text.size(-1)]).to(text.device)
 
         quantized = self.quantizer.decode(codes)
         if self.semantic_frame_rate == '25hz':
             quantized = F.interpolate(quantized, scale_factor=2, mode="nearest")##BCT
-        x, m_p, logs_p, y_mask = self.enc_p(quantized, y_lengths, text, text_lengths, ge)
+        x, m_p, logs_p, y_mask = self.enc_p(quantized, y_lengths, text, text_lengths, ge,speed)
         fea=self.bridge(x)
         fea = F.interpolate(fea, scale_factor=1.875, mode="nearest")##BCT
         ####more wn paramter to learn mel