Merge 539ad38d310d452aa1a08e659d5eac22f843a01d into c767f0b83b998e996a4d230d86da575a03f54a3f

* 修复bug

* fallbak and bug fix

GPT_SoVITS/AR/models/t2s_model.py

@@ -707,10 +707,12 @@ class Text2SemanticDecoder(nn.Module):
 
             if idx == 0:
                 attn_mask = F.pad(attn_mask[:, :, -1].unsqueeze(-2), (0, 1), value=False)
-                logits = logits[:, :-1]
             else:
                 attn_mask = F.pad(attn_mask, (0, 1), value=False)
 
+            if idx < 11:  ###至少预测出10个token不然不给停止（0.4s）
+                logits = logits[:, :-1] 
+
             samples = sample(
                 logits, y, top_k=top_k, top_p=top_p, repetition_penalty=repetition_penalty, temperature=temperature
             )[0]

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")

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(

README.md

@@ -347,7 +347,7 @@ Use v4 from v1/v2/v3 environment:
 
 2. Clone the latest codes from github.
 
-3. Download v4 pretrained models (gsv-v4-pretrained/s2v4.ckpt, and gsv-v4-pretrained/vocoder.pth) from [huggingface](https://huggingface.co/lj1995/GPT-SoVITS/tree/main) and put them into `GPT_SoVITS/pretrained_models`.
+3. Download v4 pretrained models (gsv-v4-pretrained/s2v4.pth, and gsv-v4-pretrained/vocoder.pth) from [huggingface](https://huggingface.co/lj1995/GPT-SoVITS/tree/main) and put them into `GPT_SoVITS/pretrained_models`.
 
 ## V2Pro Release Notes
 

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",

requirements.txt

@@ -19,7 +19,7 @@ torchaudio
 modelscope
 sentencepiece
 transformers>=4.43,<=4.50
-peft
+peft<0.18.0
 chardet
 PyYAML
 psutil