From abe984395cb6d8ed2055f5496d0bb26007f30365 Mon Sep 17 00:00:00 2001
From: ChasonJiang <46401978+ChasonJiang@users.noreply.github.com>
Date: Fri, 19 Dec 2025 16:05:36 +0800
Subject: [PATCH] =?UTF-8?q?=E5=AF=B9=E9=BD=90gpt=20topk=E9=BB=98=E8=AE=A4?=
 =?UTF-8?q?=E9=87=87=E6=A0=B7=E5=8F=82=E6=95=B0=20(#2696)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 GPT_SoVITS/TTS_infer_pack/TTS.py   | 4 ++--
 GPT_SoVITS/inference_webui_fast.py | 2 +-
 api_v2.py                          | 8 ++++----
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/GPT_SoVITS/TTS_infer_pack/TTS.py b/GPT_SoVITS/TTS_infer_pack/TTS.py
index 2e130978..9c8344b0 100644
--- a/GPT_SoVITS/TTS_infer_pack/TTS.py
+++ b/GPT_SoVITS/TTS_infer_pack/TTS.py
@@ -1008,7 +1008,7 @@ class TTS:
                     "aux_ref_audio_paths": [],    # list.(optional) auxiliary reference audio paths for multi-speaker tone fusion
                     "prompt_text": "",            # str.(optional) prompt text for the reference audio
                     "prompt_lang": "",            # str.(required) language of the prompt text for the reference audio
-                    "top_k": 5,                   # int. top k sampling
+                    "top_k": 15,                  # int. top k sampling
                     "top_p": 1,                   # float. top p sampling
                     "temperature": 1,             # float. temperature for sampling
                     "text_split_method": "cut1",  # str. text split method, see text_segmentation_method.py for details.
@@ -1039,7 +1039,7 @@ class TTS:
         aux_ref_audio_paths: list = inputs.get("aux_ref_audio_paths", [])
         prompt_text: str = inputs.get("prompt_text", "")
         prompt_lang: str = inputs.get("prompt_lang", "")
-        top_k: int = inputs.get("top_k", 5)
+        top_k: int = inputs.get("top_k", 15)
         top_p: float = inputs.get("top_p", 1)
         temperature: float = inputs.get("temperature", 1)
         text_split_method: str = inputs.get("text_split_method", "cut1")
diff --git a/GPT_SoVITS/inference_webui_fast.py b/GPT_SoVITS/inference_webui_fast.py
index 51a120f1..92d145b3 100644
--- a/GPT_SoVITS/inference_webui_fast.py
+++ b/GPT_SoVITS/inference_webui_fast.py
@@ -385,7 +385,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css
                         minimum=0.6, maximum=1.65, step=0.05, label="语速", value=1.0, interactive=True
                     )
                 with gr.Row():
-                    top_k = gr.Slider(minimum=1, maximum=100, step=1, label=i18n("top_k"), value=5, interactive=True)
+                    top_k = gr.Slider(minimum=1, maximum=100, step=1, label=i18n("top_k"), value=15, interactive=True)
                     top_p = gr.Slider(minimum=0, maximum=1, step=0.05, label=i18n("top_p"), value=1, interactive=True)
                 with gr.Row():
                     temperature = gr.Slider(
diff --git a/api_v2.py b/api_v2.py
index 8c83bb0f..21511db3 100644
--- a/api_v2.py
+++ b/api_v2.py
@@ -27,7 +27,7 @@ POST:
     "aux_ref_audio_paths": [],    # list.(optional) auxiliary reference audio paths for multi-speaker tone fusion
     "prompt_text": "",            # str.(optional) prompt text for the reference audio
     "prompt_lang": "",            # str.(required) language of the prompt text for the reference audio
-    "top_k": 5,                   # int. top k sampling
+    "top_k": 15,                  # int. top k sampling
     "top_p": 1,                   # float. top p sampling
     "temperature": 1,             # float. temperature for sampling
     "text_split_method": "cut5",  # str. text split method, see text_segmentation_method.py for details.
@@ -158,7 +158,7 @@ class TTS_Request(BaseModel):
     aux_ref_audio_paths: list = None
     prompt_lang: str = None
     prompt_text: str = ""
-    top_k: int = 5
+    top_k: int = 15
     top_p: float = 1
     temperature: float = 1
     text_split_method: str = "cut5"
@@ -355,7 +355,7 @@ async def tts_handle(req: dict):
                 "aux_ref_audio_paths": [],    # list.(optional) auxiliary reference audio paths for multi-speaker tone fusion
                 "prompt_text": "",            # str.(optional) prompt text for the reference audio
                 "prompt_lang": "",            # str.(required) language of the prompt text for the reference audio
-                "top_k": 5,                   # int. top k sampling
+                "top_k": 15,                  # int. top k sampling
                 "top_p": 1,                   # float. top p sampling
                 "temperature": 1,             # float. temperature for sampling
                 "text_split_method": "cut5",  # str. text split method, see text_segmentation_method.py for details.
@@ -460,7 +460,7 @@ async def tts_get_endpoint(
     aux_ref_audio_paths: list = None,
     prompt_lang: str = None,
     prompt_text: str = "",
-    top_k: int = 5,
+    top_k: int = 15,
     top_p: float = 1,
     temperature: float = 1,
     text_split_method: str = "cut5",