vits新增gpt-sovits的兼容（vits-simple-api）

config.json

@@ -475,7 +475,20 @@
     "noisew": "0.4",
     "max": "50",
     "format": "wav",
-    "sdp_radio": "0.2"
+    "sdp_radio": "0.2",
+    "gpt_sovits": {
+      "id": "0",
+      "format": "wav",
+      "lang": "auto",
+      "segment_size": "30",
+      "reference_audio": "E:\\GitHub_pro\\AI-Vtuber\\out\\gpt_sovits_67.wav",
+      "prompt_text": "所有拍到的姐妹一定不要划走",
+      "prompt_lang": "auto",
+      "preset": "default",
+      "top_k": "5",
+      "top_p": "1",
+      "temperature": "1"
+    }
   },
   "bert_vits2": {
     "type": "hiyori",

config.json.bak

@@ -475,7 +475,20 @@
     "noisew": "0.4",
     "max": "50",
     "format": "wav",
-    "sdp_radio": "0.2"
+    "sdp_radio": "0.2",
+    "gpt_sovits": {
+      "id": "0",
+      "format": "wav",
+      "lang": "auto",
+      "segment_size": "30",
+      "reference_audio": "E:\\GitHub_pro\\AI-Vtuber\\out\\gpt_sovits_67.wav",
+      "prompt_text": "所有拍到的姐妹一定不要划走",
+      "prompt_lang": "auto",
+      "preset": "default",
+      "top_k": "5",
+      "top_p": "1",
+      "temperature": "1"
+    }
   },
   "bert_vits2": {
     "type": "hiyori",

tests/test_vits/gpt-sovits.py

@@ -0,0 +1,95 @@
+import logging, asyncio, aiohttp, traceback, os
+from aiohttp import FormData
+from urllib.parse import urlencode, urljoin
+
+class TTS:
+    def __init__(self):
+        self.timeout = 60
+
+    # 请求vits_simple_api的api gpt_sovits
+    async def vits_simple_api_gpt_sovits_api(self, data):
+        try:
+            logging.debug(f"data={data}")
+            # API地址 "http://127.0.0.1:5000/voice"
+            API_URL = urljoin(data["api_ip_port"], '/voice/gpt-sovits')
+
+            data_json = {
+                "text": data["content"],
+                "id": data["id"],
+                "format": data["format"],
+                "lang": data["lang"],
+                "segment_size": data["segment_size"],
+                "prompt_text": data["prompt_text"],
+                "prompt_lang": data["prompt_lang"],
+                "preset": data["preset"],
+                "top_k": data["top_k"],
+                "top_p": data["top_p"],
+                "temperature": data["temperature"]
+            }
+
+            # 创建 FormData 对象
+            form_data = FormData()
+            # 添加文本字段
+            for key, value in data_json.items():
+                form_data.add_field(key, str(value))
+
+            # 以二进制读取模式打开音频文件，并添加到表单数据中
+            # 'reference_audio' 是字段名称，应与服务器端接收的名称一致
+            form_data.add_field('reference_audio',
+                        open(data["reference_audio"], 'rb'),
+                        content_type='audio/mpeg')  # 内容类型根据文件类型修改
+
+            logging.info(f"data_json={data_json}")
+            # logging.info(f"data={data}")
+
+            logging.info(f"API_URL={API_URL}")
+
+            # url = f"{API_URL}?{urlencode(data_json)}"
+
+            async with aiohttp.ClientSession() as session:
+                async with session.post(API_URL, data=form_data, timeout=self.timeout) as response:
+                    response = await response.read()
+                    # print(response)
+                    # file_name = 'vits_simple_api_gpt_sovits_' + self.common.get_bj_time(4) + '.wav'
+                    # voice_tmp_path = self.common.get_new_audio_path(self.audio_out_path, file_name)
+                    voice_tmp_path = '1.wav'
+                    with open(voice_tmp_path, 'wb') as f:
+                        f.write(response)
+
+                    return voice_tmp_path
+        except aiohttp.ClientError as e:
+            logging.error(traceback.format_exc())
+            logging.error(f'vits_simple_api gpt_sovits请求失败，请检查您的vits_simple_api是否启动/配置是否正确，报错内容: {e}')
+        except Exception as e:
+            logging.error(traceback.format_exc())
+            logging.error(f'vits_simple_api gpt_sovits未知错误，请检查您的vits_simple_api是否启动/配置是否正确，报错内容: {e}')
+
+        return None
+
+if __name__ == '__main__':
+    # 配置日志输出格式
+    logging.basicConfig(
+        level=logging.DEBUG,  # 设置日志级别，可以根据需求调整
+        format="%(asctime)s [%(levelname)s] %(message)s",
+        datefmt="%Y-%m-%d %H:%M:%S",
+    )
+
+    data = {
+        "api_ip_port": "http://127.0.0.1:23456/",
+        "content": "你好,你在说什么玩意，啊啊啊啊",
+        "id": 0,
+        "format": "wav",
+        "lang": "auto",
+        "segment_size": 30,
+        "reference_audio": "E:\\GitHub_pro\\AI-Vtuber\\out\\gpt_sovits_67.wav",
+        "prompt_text": "所有拍到的姐妹一定不要划走",
+        "prompt_lang": "auto",
+        "preset": "default",
+        "top_k": 5,
+        "top_p": 1,
+        "temperature": 1
+    }
+    asyncio.run(TTS().vits_simple_api_gpt_sovits_api(data))
+
+
+

utils/audio.py

@@ -660,11 +660,13 @@ async def voice_change_and_put_to_queue(message, voice_tmp_path):
                     "noisew": message["data"]["noisew"],
                     "max": message["data"]["max"],
                     "sdp_radio": message["data"]["sdp_radio"],
-                    "content": message["content"]
+                    "content": message["content"],
+                    "gpt_sovits": message["data"]["gpt_sovits"],
                 }
 
                 # 调用接口合成语音
                 voice_tmp_path = await self.my_tts.vits_api(data)
+
             elif message["tts_type"] == "bert_vits2":
                 if message["data"]["type"] == "hiyori":
                     if message["data"]["language"] == "auto":
@@ -1381,7 +1383,8 @@ async def audio_synthesis_use_local_config(self, content, audio_synthesis_type="
                 "noisew": vits["noisew"],
                 "max": vits["max"],
                 "sdp_radio": vits["sdp_radio"],
-                "content": content
+                "content": content,
+                "gpt_sovits": vits["gpt_sovits"],
             }
 
             # 调用接口合成语音

utils/audio_handle/my_tts.py

@@ -113,7 +113,69 @@ async def vits_api(self, data):
                     data_json["lang"] = "auto"
                 else:
                     data_json["lang"] = "auto"
-
+            elif data["type"] == "gpt_sovits":
+                # 请求vits_simple_api的api gpt_sovits
+                async def vits_simple_api_gpt_sovits_api(data):
+                    try:
+                        from aiohttp import FormData
+
+                        logging.debug(f"data={data}")
+                        API_URL = urljoin(data["api_ip_port"], '/voice/gpt-sovits')
+
+
+                        data_json = {
+                            "text": data["content"],
+                            "id": data["gpt_sovits"]["id"],
+                            "format": data["gpt_sovits"]["format"],
+                            "lang": data["gpt_sovits"]["lang"],
+                            "segment_size": data["gpt_sovits"]["segment_size"],
+                            "prompt_text": data["gpt_sovits"]["prompt_text"],
+                            "prompt_lang": data["gpt_sovits"]["prompt_lang"],
+                            "preset": data["gpt_sovits"]["preset"],
+                            "top_k": data["gpt_sovits"]["top_k"],
+                            "top_p": data["gpt_sovits"]["top_p"],
+                            "temperature": data["gpt_sovits"]["temperature"]
+                        }
+
+                        # 创建 FormData 对象
+                        form_data = FormData()
+                        # 添加文本字段
+                        for key, value in data_json.items():
+                            form_data.add_field(key, str(value))
+
+                        # 以二进制读取模式打开音频文件，并添加到表单数据中
+                        # 'reference_audio' 是字段名称，应与服务器端接收的名称一致
+                        form_data.add_field('reference_audio',
+                                    open(data["gpt_sovits"]["reference_audio"], 'rb'),
+                                    content_type='audio/mpeg')  # 内容类型根据文件类型修改
+
+                        logging.debug(f"data_json={data_json}")
+
+                        logging.debug(f"API_URL={API_URL}")
+
+                        async with aiohttp.ClientSession() as session:
+                            async with session.post(API_URL, data=form_data, timeout=self.timeout) as response:
+                                response = await response.read()
+                                # print(response)
+                                file_name = 'vits_simple_api_' + self.common.get_bj_time(4) + '.wav'
+                                voice_tmp_path = self.common.get_new_audio_path(self.audio_out_path, file_name)
+
+                                with open(voice_tmp_path, 'wb') as f:
+                                    f.write(response)
+
+                                return voice_tmp_path
+                    except aiohttp.ClientError as e:
+                        logging.error(traceback.format_exc())
+                        logging.error(f'vits_simple_api gpt_sovits请求失败，请检查您的vits_simple_api是否启动/配置是否正确，报错内容: {e}')
+                    except Exception as e:
+                        logging.error(traceback.format_exc())
+                        logging.error(f'vits_simple_api gpt_sovits未知错误，请检查您的vits_simple_api是否启动/配置是否正确，报错内容: {e}')
+
+                    return None
+
+                voice_tmp_path = await vits_simple_api_gpt_sovits_api(data)
+                return voice_tmp_path
+
             # logging.info(f"data_json={data_json}")
             # logging.info(f"data={data}")
 
@@ -824,3 +886,5 @@ def replace_empty_strings_with_none(input_dict):
 
         return None
 
+
+

webui.py

@@ -1566,6 +1566,18 @@ def common_textarea_handle(content):
                     config_data["vits"]["format"] = input_vits_format.value
                     config_data["vits"]["sdp_radio"] = input_vits_sdp_radio.value
 
+                    config_data["vits"]["gpt_sovits"]["id"] = input_vits_gpt_sovits_id.value
+                    config_data["vits"]["gpt_sovits"]["lang"] = select_vits_gpt_sovits_lang.value
+                    config_data["vits"]["gpt_sovits"]["format"] = input_vits_gpt_sovits_format.value
+                    config_data["vits"]["gpt_sovits"]["segment_size"] = input_vits_gpt_sovits_segment_size.value
+                    config_data["vits"]["gpt_sovits"]["reference_audio"] = input_vits_gpt_sovits_reference_audio.value
+                    config_data["vits"]["gpt_sovits"]["prompt_text"] = input_vits_gpt_sovits_prompt_text.value
+                    config_data["vits"]["gpt_sovits"]["prompt_lang"] = select_vits_gpt_sovits_prompt_lang.value
+                    config_data["vits"]["gpt_sovits"]["top_k"] = input_vits_gpt_sovits_top_k.value
+                    config_data["vits"]["gpt_sovits"]["top_p"] = input_vits_gpt_sovits_top_p.value
+                    config_data["vits"]["gpt_sovits"]["temperature"] = input_vits_gpt_sovits_temperature.value
+                    config_data["vits"]["gpt_sovits"]["preset"] = input_vits_gpt_sovits_preset.value
+
                 if config.get("webui", "show_card", "tts", "bert_vits2"):
                     config_data["bert_vits2"]["type"] = select_bert_vits2_type.value
                     config_data["bert_vits2"]["api_ip_port"] = input_bert_vits2_api_ip_port.value
@@ -3356,11 +3368,11 @@ def clear_tts_common_audio_card(file_path):
 
             if config.get("webui", "show_card", "tts", "vits"):
                 with ui.card().style(card_css):
-                    ui.label("VITS")
+                    ui.label("VITS-Simple-API")
                     with ui.row():
                         select_vits_type = ui.select(
                             label='类型', 
-                            options={'vits': 'vits', 'bert_vits2': 'bert_vits2'}, 
+                            options={'vits': 'vits', 'bert_vits2': 'bert_vits2', 'gpt_sovits': 'gpt_sovits'}, 
                             value=config.get("vits", "type")
                         ).style("width:200px;")
                         input_vits_config_path = ui.input(label='配置文件路径', placeholder='模型配置文件存储路径', value=config.get("vits", "config_path")).style("width:200px;")
@@ -3385,7 +3397,32 @@ def clear_tts_common_audio_card(file_path):
                         input_vits_format = ui.input(label='音频格式', placeholder='支持wav,ogg,silk,mp3,flac', value=config.get("vits", "format")).style("width:200px;")
 
                         input_vits_sdp_radio = ui.input(label='SDP/DP混合比', placeholder='SDP/DP混合比：SDP在合成时的占比，理论上此比率越高，合成的语音语调方差越大。', value=config.get("vits", "sdp_radio")).style("width:200px;")
-
+
+                    with ui.expansion('GPT-SOVITS', icon="settings", value=True).classes('w-full'):
+                        with ui.row():
+                            input_vits_gpt_sovits_id = ui.input(label='说话人ID', value=config.get("vits", "gpt_sovits", "id"), placeholder='API启动时会给配置文件重新划分id，一般为拼音顺序排列，从0开始').style("width:100px;")
+
+                            select_vits_gpt_sovits_lang = ui.select(
+                                label='语言', 
+                                options={'auto': '自动', 'zh': '中文', 'jp': '英文', 'en': '日文'}, 
+                                value=config.get("vits", "gpt_sovits", "lang")
+                            ).style("width:100px;")
+                            input_vits_gpt_sovits_format = ui.input(label='音频格式', value=config.get("vits", "gpt_sovits", "format"), placeholder='支持wav,ogg,silk,mp3,flac').style("width:100px;")
+                            input_vits_gpt_sovits_segment_size = ui.input(label='segment_size', value=config.get("vits", "gpt_sovits", "segment_size"), placeholder='segment_size').style("width:100px;")
+                            input_vits_gpt_sovits_reference_audio = ui.input(label='参考音频路径', value=config.get("vits", "gpt_sovits", "reference_audio"), placeholder='参考音频路径').style("width:200px;")
+                            input_vits_gpt_sovits_prompt_text = ui.input(label='参考音频文本内容', value=config.get("vits", "gpt_sovits", "prompt_text"), placeholder='参考音频文本内容').style("width:200px;")
+                            select_vits_gpt_sovits_prompt_lang = ui.select(
+                                label='参考音频语言', 
+                                options={'auto': '自动', 'zh': '中文', 'jp': '英文', 'en': '日文'}, 
+                                value=config.get("vits", "gpt_sovits", "prompt_lang")
+                            ).style("width:150px;")
+                        with ui.row():
+                            input_vits_gpt_sovits_top_k = ui.input(label='top_k', value=config.get("vits", "gpt_sovits", "top_k"), placeholder='top_k').style("width:100px;")
+                            input_vits_gpt_sovits_top_p = ui.input(label='top_p', value=config.get("vits", "gpt_sovits", "top_p"), placeholder='top_p').style("width:100px;")
+                            input_vits_gpt_sovits_temperature = ui.input(label='temperature', value=config.get("vits", "gpt_sovits", "temperature"), placeholder='temperature').style("width:100px;")
+                            input_vits_gpt_sovits_preset = ui.input(label='preset', value=config.get("vits", "gpt_sovits", "preset"), placeholder='preset').style("width:100px;")
+
+
             if config.get("webui", "show_card", "tts", "bert_vits2"):
                 with ui.card().style(card_css):
                     ui.label("bert_vits2")