{"id":128,"date":"2025-02-12T11:11:00","date_gmt":"2025-02-12T03:11:00","guid":{"rendered":"http:\/\/blog.sayoung.xyz\/?p=128"},"modified":"2026-02-10T15:27:55","modified_gmt":"2026-02-10T07:27:55","slug":"%e5%ad%a6%e4%b9%a0%e6%97%a5%e5%bf%97%e7%bb%93%e5%90%88gpt-sovites%e4%b8%8ellm%ef%bc%881%ef%bc%89","status":"publish","type":"post","link":"https:\/\/blog.sayoung.wang\/index.php\/2025\/02\/12\/%e5%ad%a6%e4%b9%a0%e6%97%a5%e5%bf%97%e7%bb%93%e5%90%88gpt-sovites%e4%b8%8ellm%ef%bc%881%ef%bc%89\/","title":{"rendered":"\u5b66\u4e60\u65e5\u5fd7|\u7ed3\u5408GPT-SoVites\u4e0eLLM\uff081\uff09"},"content":{"rendered":"\n

\u6700\u8fd1\u6362\u4e86\u7535\u8111\uff0c\u7ec8\u4e8e\u53ef\u4ee5\u8bd5\u8bd5GSV\u548cRVC\u4e86\u3002\u5728\u7b80\u5355\u4f53\u9a8c\u4e00\u756a\u540e\uff0c\u840c\u751f\u4e86\u5c06\u5176\u4e0eLLM\u7ed3\u5408\u8d77\u6765\u7684\u60f3\u6cd5\u3002 \uff08\u6bd5\u7adf\u8c01\u4e0d\u60f3\u8ba9AI\u7528\u81ea\u5df1\u559c\u7231\u7684\u865a\u62df\u89d2\u8272\u7684\u58f0\u97f3\u8bf4\u8bdd\u5462<\/hm><\/p>\n\n\n\n

\u5b9e\u73b0\u65b9\u6cd5\u4e5f\u7b80\u5355\uff0c\u5b98\u65b9\u5df2\u7ecf\u63d0\u4f9b\u4e86API\uff0c\u53ea\u9700\u8981\u628aLLM\u8f93\u51fa\u7684\u8bdd\u7528\u5b83\u5904\u7406\u4e00\u904d\u5c31\u597d\u4e86\u3002\uff08\u5f53\u7136\uff0c\u8003\u8651\u5230TTS\u9700\u8981\u6d88\u8017\u4e00\u5b9a\u65f6\u95f4\uff0c\u8fd8\u9700\u8981\u5bf9\u53e5\u5b50\u8fdb\u884c\u62c6\u5206\uff0c\u5206\u6bb5\u8fdb\u884c\u6d41\u5f0f\u8f93\u51fa\u3002\uff09<\/p>\n\n\n\n

API<\/h2>\n\n\n\n

\u4ee5v2\u7248\u672c\u4e3a\u4f8b\uff0c\u8981\u60f3\u4f7f\u7528GSV\u7684API\uff0c\u9996\u5148\u8981\u8fd0\u884c\u5176\u6839\u76ee\u5f55\u4e0b\u7684api_v2.py\u6587\u4ef6\u3002\u9700\u8981\u5728\u6839\u76ee\u5f55\u4e0b\u6253\u5f00CMD\uff0c\u7136\u540e\u6267\u884c\u4ee5\u4e0b\u547d\u4ee4\u3002<\/p>\n\n\n

\nruntime\\python api_v2.py -a 127.0.0.1 -p 9880 -c GPT_SoVITS\/configs\/tts_infer.yaml\n<\/pre><\/div>\n\n\n
    <\/ul>\n\n\n\n
      \n
    • -a<\/strong> – \u7ed1\u5b9a\u5730\u5740, \u9ed8\u8ba4”127.0.0.1″<\/li>\n<\/ul>\n\n\n\n
        \n
      • -p<\/strong> – \u7ed1\u5b9a\u7aef\u53e3, \u9ed8\u8ba49880<\/li>\n<\/ul>\n\n\n\n
          \n
        • -c<\/strong>– TTS\u914d\u7f6e\u6587\u4ef6\u8def\u5f84, \u9ed8\u8ba4”GPT_SoVITS\/configs\/tts_infer.yaml”<\/li>\n<\/ul>\n\n\n\n
          \u5b98\u65b9\u63d0\u4f9b\u7684API\u652f\u6301GET\/POST\u65b9\u6cd5<\/p>\n\n\n
          \n#GET:\nhttp:\/\/127.0.0.1:9880\/tts?text=\u5148\u5e1d\u521b\u4e1a\u672a\u534a\u800c\u4e2d\u9053\u5d29\u6b82\uff0c\u4eca\u5929\u4e0b\u4e09\u5206\uff0c\u76ca\u5dde\u75b2\u5f0a\uff0c\u6b64\u8bda\u5371\u6025\u5b58\u4ea1\u4e4b\u79cb\u4e5f\u3002&text_lang=zh&ref_audio_path=archive_jingyuan_1.wav&prompt_lang=zh&prompt_text=\u6211\u662f\u300c\u7f57\u6d6e\u300d\u4e91\u9a91\u5c06\u519b\u666f\u5143\u3002\u4e0d\u5fc5\u62d8\u8c28\uff0c\u300c\u5c06\u519b\u300d\u53ea\u662f\u4e00\u65f6\u7684\u8eab\u4efd\uff0c\u4f60\u79f0\u547c\u6211\u666f\u5143\u4fbf\u53ef&text_split_method=cut5&batch_size=1&media_type=wav&streaming_mode=true\n<\/pre><\/div>\n\n
          \n#POST\n{\n    "text": "",                   # str.(required) text to be synthesized\n    "text_lang: "",               # str.(required) language of the text to be synthesized\n    "ref_audio_path": "",         # str.(required) reference audio path\n    "aux_ref_audio_paths": [],    # list.(optional) auxiliary reference audio paths for multi-speaker synthesis\n    "prompt_text": "",            # str.(optional) prompt text for the reference audio\n    "prompt_lang": "",            # str.(required) language of the prompt text for the reference audio\n    "top_k": 5,                   # int. top k sampling\n    "top_p": 1,                   # float. top p sampling\n    "temperature": 1,             # float. temperature for sampling\n    "text_split_method": "cut0",  # str. text split method, see text_segmentation_method.py for details.\n    "batch_size": 1,              # int. batch size for inference\n    "batch_threshold": 0.75,      # float. threshold for batch splitting.\n    "split_bucket": True,          # bool. whether to split the batch into multiple buckets.\n    "return_fragment": False,     # bool. step by step return the audio fragment.\n    "speed_factor":1.0,           # float. control the speed of the synthesized audio.\n    "streaming_mode": False,      # bool. whether to return a streaming response.\n    "seed": -1,                   # int. random seed for reproducibility.\n    "parallel_infer": True,       # bool. whether to use parallel inference.\n    "repetition_penalty": 1.35    # float. repetition penalty for T2S model.\n}\n<\/pre><\/div>\n\n\n
          \u793a\u4f8b\u4ee3\u7801\u5982\u4e0b\uff08\u4f7f\u7528\u9ed8\u8ba4\u914d\u7f6e\uff09\uff1a<\/p>\n\n\n
          \nimport requests\nimport json\n\nurl = "http:\/\/127.0.0.1:9880\/tts"\n\n    data = {\n        "text": "\u6211\u5728\u8fd9\u91cc\uff0c\u535a\u58eb\u3002",                   \n        "text_lang": "zh",               \n        "ref_audio_path": "\u5e72\u5458\u62a5\u5230.wav",        \n        "aux_ref_audio_paths": [],   \n        "prompt_text": "\u661f\u8c61\u5b66\u8005\uff0c\u661f\u6781\uff0c\u4ee5\u8fd1\u536b\u5e72\u5458\u8eab\u4efd\u4efb\u804c\uff0c\u4eca\u540e\u5c31\u7531\u60a8\u5dee\u9063\u4e86\uff0c\u535a\u58eb\u3002",      \n        "prompt_lang": "zh",            \n        "top_k": 5,                   \n        "top_p": 1,                   \n        "temperature": 1,            \n        "text_split_method": "cut5",  \n        "batch_size": 1,              \n        "batch_threshold": 0.75,      \n        "split_bucket": True,          \n        "speed_factor":1.0,           \n        "fragment_interval":0.3,      \n        "seed": -1,                   \n        "media_type": "wav",          \n        "streaming_mode": False,      \n        "parallel_infer": True,       \n        "repetition_penalty": 1.35    \n    }\n\n    headers = {"Content-Type": "application\/json"}\n\n    response = requests.post(url, data=json.dumps(data), headers=headers)\n\n    if response.status_code == 200:\n        # \u4fdd\u5b58\u751f\u6210\u7684\u97f3\u9891\n        with open("output.wav", "wb") as f:\n            f.write(response.content) \n        print("\u97f3\u9891\u751f\u6210\u6210\u529f\uff01")\n        return response.content\n    else:\n        print(f"\u8bf7\u6c42\u5931\u8d25\uff0c\u72b6\u6001\u7801\uff1a{response.status_code}, \u9519\u8bef\u4fe1\u606f\uff1a{response.text}")\n<\/pre><\/div>\n\n\n
          \u6548\u679c\u5982\u4e0b\uff1a<\/p>\n\n\n\n