Update gui.py

1. Fix the issue that inference_gui needs to reload models for each inference.
2. Simplify GUI's code and address various inefficiencies, including: enabling direct input of ref text and target text (akin to the WebUI), facilitating file selection for ref audio uploads, adding language options for CH-EN/JA-EN/Multi (with Multi as the default), standardizing variable name to enhance readability.

GPT_SoVITS/inference_gui.py

@@ -1,22 +1,592 @@
+'''
+按中英混合识别
+按日英混合识别
+多语种启动切分识别语种
+全部按中文识别
+全部按英文识别
+全部按日文识别
+'''
+import os, re, logging
+import LangSegment
+logging.getLogger("markdown_it").setLevel(logging.ERROR)
+logging.getLogger("urllib3").setLevel(logging.ERROR)
+logging.getLogger("httpcore").setLevel(logging.ERROR)
+logging.getLogger("httpx").setLevel(logging.ERROR)
+logging.getLogger("asyncio").setLevel(logging.ERROR)
+logging.getLogger("charset_normalizer").setLevel(logging.ERROR)
+logging.getLogger("torchaudio._extension").setLevel(logging.ERROR)
+import pdb
+import torch
+
+if os.path.exists("./gweight.txt"):
+    with open("./gweight.txt", 'r', encoding="utf-8") as file:
+        gweight_data = file.read()
+        gpt_path = os.environ.get(
+            "gpt_path", gweight_data)
+else:
+    gpt_path = os.environ.get(
+        "gpt_path", "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt")
+
+if os.path.exists("./sweight.txt"):
+    with open("./sweight.txt", 'r', encoding="utf-8") as file:
+        sweight_data = file.read()
+        sovits_path = os.environ.get("sovits_path", sweight_data)
+else:
+    sovits_path = os.environ.get("sovits_path", "GPT_SoVITS/pretrained_models/s2G488k.pth")
+# gpt_path = os.environ.get(
+#     "gpt_path", "pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"
+# )
+# sovits_path = os.environ.get("sovits_path", "pretrained_models/s2G488k.pth")
+cnhubert_base_path = os.environ.get(
+    "cnhubert_base_path", "GPT_SoVITS/pretrained_models/chinese-hubert-base"
+)
+bert_path = os.environ.get(
+    "bert_path", "GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large"
+)
+infer_ttswebui = os.environ.get("infer_ttswebui", 9872)
+infer_ttswebui = int(infer_ttswebui)
+is_share = os.environ.get("is_share", "False")
+is_share = eval(is_share)
+if "_CUDA_VISIBLE_DEVICES" in os.environ:
+    os.environ["CUDA_VISIBLE_DEVICES"] = os.environ["_CUDA_VISIBLE_DEVICES"]
+is_half = eval(os.environ.get("is_half", "True")) and torch.cuda.is_available()
+punctuation = set(['!', '?', '…', ',', '.', '-'," "])
+
 import sys
 from PyQt5.QtCore import QEvent
 from PyQt5.QtWidgets import QApplication, QMainWindow, QLabel, QLineEdit, QPushButton, QTextEdit
 from PyQt5.QtWidgets import QGridLayout, QVBoxLayout, QWidget, QFileDialog, QStatusBar, QComboBox
 import soundfile as sf
+from transformers import AutoModelForMaskedLM, AutoTokenizer
+import numpy as np
+import librosa
+from feature_extractor import cnhubert
 
+cnhubert.cnhubert_base_path = cnhubert_base_path
+
+from module.models import SynthesizerTrn
+from AR.models.t2s_lightning_module import Text2SemanticLightningModule
+from text import cleaned_text_to_sequence
+from text.cleaner import clean_text
+from time import time as ttime
+from module.mel_processing import spectrogram_torch
+from my_utils import load_audio
 from tools.i18n.i18n import I18nAuto
+
 i18n = I18nAuto()
 
-from GPT_SoVITS.inference_webui import change_gpt_weights, change_sovits_weights, get_tts_wav
+# os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1'  # 确保直接启动推理UI时也能够设置。
+
+if torch.cuda.is_available():
+    device = "cuda"
+else:
+    device = "cpu"
+
+tokenizer = AutoTokenizer.from_pretrained(bert_path)
+bert_model = AutoModelForMaskedLM.from_pretrained(bert_path)
+if is_half == True:
+    bert_model = bert_model.half().to(device)
+else:
+    bert_model = bert_model.to(device)
+
+
+def get_bert_feature(text, word2ph):
+    with torch.no_grad():
+        inputs = tokenizer(text, return_tensors="pt")
+        for i in inputs:
+            inputs[i] = inputs[i].to(device)
+        res = bert_model(**inputs, output_hidden_states=True)
+        res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()[1:-1]
+    assert len(word2ph) == len(text)
+    phone_level_feature = []
+    for i in range(len(word2ph)):
+        repeat_feature = res[i].repeat(word2ph[i], 1)
+        phone_level_feature.append(repeat_feature)
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+    return phone_level_feature.T
+
+
+class DictToAttrRecursive(dict):
+    def __init__(self, input_dict):
+        super().__init__(input_dict)
+        for key, value in input_dict.items():
+            if isinstance(value, dict):
+                value = DictToAttrRecursive(value)
+            self[key] = value
+            setattr(self, key, value)
+
+    def __getattr__(self, item):
+        try:
+            return self[item]
+        except KeyError:
+            raise AttributeError(f"Attribute {item} not found")
+
+    def __setattr__(self, key, value):
+        if isinstance(value, dict):
+            value = DictToAttrRecursive(value)
+        super(DictToAttrRecursive, self).__setitem__(key, value)
+        super().__setattr__(key, value)
+
+    def __delattr__(self, item):
+        try:
+            del self[item]
+        except KeyError:
+            raise AttributeError(f"Attribute {item} not found")
+
+
+ssl_model = cnhubert.get_model()
+if is_half == True:
+    ssl_model = ssl_model.half().to(device)
+else:
+    ssl_model = ssl_model.to(device)
+
+
+def change_sovits_weights(sovits_path):
+    global vq_model, hps
+    dict_s2 = torch.load(sovits_path, map_location="cpu")
+    hps = dict_s2["config"]
+    hps = DictToAttrRecursive(hps)
+    hps.model.semantic_frame_rate = "25hz"
+    vq_model = SynthesizerTrn(
+        hps.data.filter_length // 2 + 1,
+        hps.train.segment_size // hps.data.hop_length,
+        n_speakers=hps.data.n_speakers,
+        **hps.model
+    )
+    if ("pretrained" not in sovits_path):
+        del vq_model.enc_q
+    if is_half == True:
+        vq_model = vq_model.half().to(device)
+    else:
+        vq_model = vq_model.to(device)
+    vq_model.eval()
+    print(vq_model.load_state_dict(dict_s2["weight"], strict=False))
+    with open("./sweight.txt", "w", encoding="utf-8") as f:
+        f.write(sovits_path)
+
+
+change_sovits_weights(sovits_path)
+
+
+def change_gpt_weights(gpt_path):
+    global hz, max_sec, t2s_model, config
+    hz = 50
+    dict_s1 = torch.load(gpt_path, map_location="cpu")
+    config = dict_s1["config"]
+    max_sec = config["data"]["max_sec"]
+    t2s_model = Text2SemanticLightningModule(config, "****", is_train=False)
+    t2s_model.load_state_dict(dict_s1["weight"])
+    if is_half == True:
+        t2s_model = t2s_model.half()
+    t2s_model = t2s_model.to(device)
+    t2s_model.eval()
+    total = sum([param.nelement() for param in t2s_model.parameters()])
+    print("Number of parameter: %.2fM" % (total / 1e6))
+    with open("./gweight.txt", "w", encoding="utf-8") as f: f.write(gpt_path)
+
+
+change_gpt_weights(gpt_path)
+
+
+def get_spepc(hps, filename):
+    audio = load_audio(filename, int(hps.data.sampling_rate))
+    audio = torch.FloatTensor(audio)
+    audio_norm = audio
+    audio_norm = audio_norm.unsqueeze(0)
+    spec = spectrogram_torch(
+        audio_norm,
+        hps.data.filter_length,
+        hps.data.sampling_rate,
+        hps.data.hop_length,
+        hps.data.win_length,
+        center=False,
+    )
+    return spec
+
+
+dict_language = {
+    i18n("中文"): "all_zh",#全部按中文识别
+    i18n("英文"): "en",#全部按英文识别#######不变
+    i18n("日文"): "all_ja",#全部按日文识别
+    i18n("中英混合"): "zh",#按中英混合识别####不变
+    i18n("日英混合"): "ja",#按日英混合识别####不变
+    i18n("多语种混合"): "auto",#多语种启动切分识别语种
+}
+
+
+def clean_text_inf(text, language):
+    phones, word2ph, norm_text = clean_text(text, language)
+    phones = cleaned_text_to_sequence(phones)
+    return phones, word2ph, norm_text
+
+dtype=torch.float16 if is_half == True else torch.float32
+def get_bert_inf(phones, word2ph, norm_text, language):
+    language=language.replace("all_","")
+    if language == "zh":
+        bert = get_bert_feature(norm_text, word2ph).to(device)#.to(dtype)
+    else:
+        bert = torch.zeros(
+            (1024, len(phones)),
+            dtype=torch.float16 if is_half == True else torch.float32,
+        ).to(device)
+
+    return bert
+
+
+splits = {"，", "。", "？", "！", ",", ".", "?", "!", "~", ":", "：", "—", "…", }
+
+
+def get_first(text):
+    pattern = "[" + "".join(re.escape(sep) for sep in splits) + "]"
+    text = re.split(pattern, text)[0].strip()
+    return text
+
+
+def get_phones_and_bert(text,language):
+    if language in {"en","all_zh","all_ja"}:
+        language = language.replace("all_","")
+        if language == "en":
+            LangSegment.setfilters(["en"])
+            formattext = " ".join(tmp["text"] for tmp in LangSegment.getTexts(text))
+        else:
+            # 因无法区别中日文汉字,以用户输入为准
+            formattext = text
+        while "  " in formattext:
+            formattext = formattext.replace("  ", " ")
+        phones, word2ph, norm_text = clean_text_inf(formattext, language)
+        if language == "zh":
+            bert = get_bert_feature(norm_text, word2ph).to(device)
+        else:
+            bert = torch.zeros(
+                (1024, len(phones)),
+                dtype=torch.float16 if is_half == True else torch.float32,
+            ).to(device)
+    elif language in {"zh", "ja","auto"}:
+        textlist=[]
+        langlist=[]
+        LangSegment.setfilters(["zh","ja","en","ko"])
+        if language == "auto":
+            for tmp in LangSegment.getTexts(text):
+                if tmp["lang"] == "ko":
+                    langlist.append("zh")
+                    textlist.append(tmp["text"])
+                else:
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+        else:
+            for tmp in LangSegment.getTexts(text):
+                if tmp["lang"] == "en":
+                    langlist.append(tmp["lang"])
+                else:
+                    # 因无法区别中日文汉字,以用户输入为准
+                    langlist.append(language)
+                textlist.append(tmp["text"])
+        print(textlist)
+        print(langlist)
+        phones_list = []
+        bert_list = []
+        norm_text_list = []
+        for i in range(len(textlist)):
+            lang = langlist[i]
+            phones, word2ph, norm_text = clean_text_inf(textlist[i], lang)
+            bert = get_bert_inf(phones, word2ph, norm_text, lang)
+            phones_list.append(phones)
+            norm_text_list.append(norm_text)
+            bert_list.append(bert)
+        bert = torch.cat(bert_list, dim=1)
+        phones = sum(phones_list, [])
+        norm_text = ''.join(norm_text_list)
+
+    return phones,bert.to(dtype),norm_text
+
+
+def merge_short_text_in_array(texts, threshold):
+    if (len(texts)) < 2:
+        return texts
+    result = []
+    text = ""
+    for ele in texts:
+        text += ele
+        if len(text) >= threshold:
+            result.append(text)
+            text = ""
+    if (len(text) > 0):
+        if len(result) == 0:
+            result.append(text)
+        else:
+            result[len(result) - 1] += text
+    return result
+
+def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切"), top_k=20, top_p=0.6, temperature=0.6, ref_free = False):
+    if prompt_text is None or len(prompt_text) == 0:
+        ref_free = True
+    t0 = ttime()
+    prompt_language = dict_language[prompt_language]
+    text_language = dict_language[text_language]
+    if not ref_free:
+        prompt_text = prompt_text.strip("\n")
+        if (prompt_text[-1] not in splits): prompt_text += "。" if prompt_language != "en" else "."
+        print(i18n("实际输入的参考文本:"), prompt_text)
+    text = text.strip("\n")
+    text = replace_consecutive_punctuation(text)
+    if (text[0] not in splits and len(get_first(text)) < 4): text = "。" + text if text_language != "en" else "." + text
+    
+    print(i18n("实际输入的目标文本:"), text)
+    zero_wav = np.zeros(
+        int(hps.data.sampling_rate * 0.3),
+        dtype=np.float16 if is_half == True else np.float32,
+    )
+    with torch.no_grad():
+        wav16k, sr = librosa.load(ref_wav_path, sr=16000)
+        if (wav16k.shape[0] > 160000 or wav16k.shape[0] < 48000):
+            raise OSError(i18n("参考音频在3~10秒范围外，请更换！"))
+        wav16k = torch.from_numpy(wav16k)
+        zero_wav_torch = torch.from_numpy(zero_wav)
+        if is_half == True:
+            wav16k = wav16k.half().to(device)
+            zero_wav_torch = zero_wav_torch.half().to(device)
+        else:
+            wav16k = wav16k.to(device)
+            zero_wav_torch = zero_wav_torch.to(device)
+        wav16k = torch.cat([wav16k, zero_wav_torch])
+        ssl_content = ssl_model.model(wav16k.unsqueeze(0))[
+            "last_hidden_state"
+        ].transpose(
+            1, 2
+        )  # .float()
+        codes = vq_model.extract_latent(ssl_content)
+   
+        prompt_semantic = codes[0, 0]
+    t1 = ttime()
+
+    if (how_to_cut == i18n("凑四句一切")):
+        text = cut1(text)
+    elif (how_to_cut == i18n("凑50字一切")):
+        text = cut2(text)
+    elif (how_to_cut == i18n("按中文句号。切")):
+        text = cut3(text)
+    elif (how_to_cut == i18n("按英文句号.切")):
+        text = cut4(text)
+    elif (how_to_cut == i18n("按标点符号切")):
+        text = cut5(text)
+    while "\n\n" in text:
+        text = text.replace("\n\n", "\n")
+    print(i18n("实际输入的目标文本(切句后):"), text)
+    texts = text.split("\n")
+    texts = process_text(texts)
+    texts = merge_short_text_in_array(texts, 5)
+    audio_opt = []
+    if not ref_free:
+        phones1,bert1,norm_text1=get_phones_and_bert(prompt_text, prompt_language)
+
+    for text in texts:
+        # 解决输入目标文本的空行导致报错的问题
+        if (len(text.strip()) == 0):
+            continue
+        if (text[-1] not in splits): text += "。" if text_language != "en" else "."
+        print(i18n("实际输入的目标文本(每句):"), text)
+        phones2,bert2,norm_text2=get_phones_and_bert(text, text_language)
+        print(i18n("前端处理后的文本(每句):"), norm_text2)
+        if not ref_free:
+            bert = torch.cat([bert1, bert2], 1)
+            all_phoneme_ids = torch.LongTensor(phones1+phones2).to(device).unsqueeze(0)
+        else:
+            bert = bert2
+            all_phoneme_ids = torch.LongTensor(phones2).to(device).unsqueeze(0)
+
+        bert = bert.to(device).unsqueeze(0)
+        all_phoneme_len = torch.tensor([all_phoneme_ids.shape[-1]]).to(device)
+        prompt = prompt_semantic.unsqueeze(0).to(device)
+        t2 = ttime()
+        with torch.no_grad():
+            # pred_semantic = t2s_model.model.infer(
+            pred_semantic, idx = t2s_model.model.infer_panel(
+                all_phoneme_ids,
+                all_phoneme_len,
+                None if ref_free else prompt,
+                bert,
+                # prompt_phone_len=ph_offset,
+                top_k=top_k,
+                top_p=top_p,
+                temperature=temperature,
+                early_stop_num=hz * max_sec,
+            )
+        t3 = ttime()
+        # print(pred_semantic.shape,idx)
+        pred_semantic = pred_semantic[:, -idx:].unsqueeze(
+            0
+        )  # .unsqueeze(0)#mq要多unsqueeze一次
+        refer = get_spepc(hps, ref_wav_path)  # .to(device)
+        if is_half == True:
+            refer = refer.half().to(device)
+        else:
+            refer = refer.to(device)
+        # audio = vq_model.decode(pred_semantic, all_phoneme_ids, refer).detach().cpu().numpy()[0, 0]
+        audio = (
+            vq_model.decode(
+                pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refer
+            )
+                .detach()
+                .cpu()
+                .numpy()[0, 0]
+        )  ###试试重建不带上prompt部分
+        max_audio=np.abs(audio).max()#简单防止16bit爆音
+        if max_audio>1:audio/=max_audio
+        audio_opt.append(audio)
+        audio_opt.append(zero_wav)
+        t4 = ttime()
+    print("%.3f\t%.3f\t%.3f\t%.3f" % (t1 - t0, t2 - t1, t3 - t2, t4 - t3))
+    yield hps.data.sampling_rate, (np.concatenate(audio_opt, 0) * 32768).astype(
+        np.int16
+    )
+
+
+def split(todo_text):
+    todo_text = todo_text.replace("……", "。").replace("——", "，")
+    if todo_text[-1] not in splits:
+        todo_text += "。"
+    i_split_head = i_split_tail = 0
+    len_text = len(todo_text)
+    todo_texts = []
+    while 1:
+        if i_split_head >= len_text:
+            break  # 结尾一定有标点，所以直接跳出即可，最后一段在上次已加入
+        if todo_text[i_split_head] in splits:
+            i_split_head += 1
+            todo_texts.append(todo_text[i_split_tail:i_split_head])
+            i_split_tail = i_split_head
+        else:
+            i_split_head += 1
+    return todo_texts
+
+
+def cut1(inp):
+    inp = inp.strip("\n")
+    inps = split(inp)
+    split_idx = list(range(0, len(inps), 4))
+    split_idx[-1] = None
+    if len(split_idx) > 1:
+        opts = []
+        for idx in range(len(split_idx) - 1):
+            opts.append("".join(inps[split_idx[idx]: split_idx[idx + 1]]))
+    else:
+        opts = [inp]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return "\n".join(opts)
+
+
+def cut2(inp):
+    inp = inp.strip("\n")
+    inps = split(inp)
+    if len(inps) < 2:
+        return inp
+    opts = []
+    summ = 0
+    tmp_str = ""
+    for i in range(len(inps)):
+        summ += len(inps[i])
+        tmp_str += inps[i]
+        if summ > 50:
+            summ = 0
+            opts.append(tmp_str)
+            tmp_str = ""
+    if tmp_str != "":
+        opts.append(tmp_str)
+    # print(opts)
+    if len(opts) > 1 and len(opts[-1]) < 50:  ##如果最后一个太短了，和前一个合一起
+        opts[-2] = opts[-2] + opts[-1]
+        opts = opts[:-1]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return "\n".join(opts)
+
+
+def cut3(inp):
+    inp = inp.strip("\n")
+    opts = ["%s" % item for item in inp.strip("。").split("。")]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return  "\n".join(opts)
+
+def cut4(inp):
+    inp = inp.strip("\n")
+    opts = ["%s" % item for item in inp.strip(".").split(".")]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return "\n".join(opts)
+
+
+# contributed by https://github.com/AI-Hobbyist/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py
+def cut5(inp):
+    # if not re.search(r'[^\w\s]', inp[-1]):
+    # inp += '。'
+    inp = inp.strip("\n")
+    punds = r'[,.;?!、，。？！;：…]'
+    items = re.split(f'({punds})', inp)
+    mergeitems = ["".join(group) for group in zip(items[::2], items[1::2])]
+    # 在句子不存在符号或句尾无符号的时候保证文本完整
+    if len(items)%2 == 1:
+        mergeitems.append(items[-1])
+    opt = [item for item in mergeitems if not set(item).issubset(punctuation)]
+    return "\n".join(opt)
+
+
+def custom_sort_key(s):
+    # 使用正则表达式提取字符串中的数字部分和非数字部分
+    parts = re.split('(\d+)', s)
+    # 将数字部分转换为整数，非数字部分保持不变
+    parts = [int(part) if part.isdigit() else part for part in parts]
+    return parts
+
+def process_text(texts):
+    _text=[]
+    if all(text in [None, " ", "\n",""] for text in texts):
+        raise ValueError(i18n("请输入有效文本"))
+    for text in texts:
+        if text in  [None, " ", ""]:
+            pass
+        else:
+            _text.append(text)
+    return _text
+
+
+def replace_consecutive_punctuation(text):
+    punctuations = ''.join(re.escape(p) for p in punctuation)
+    pattern = f'([{punctuations}])([{punctuations}])+'
+    result = re.sub(pattern, r'\1', text)
+    return result
+
+
+def change_choices():
+    SoVITS_names, GPT_names = get_weights_names()
+    return {"choices": sorted(SoVITS_names, key=custom_sort_key), "__type__": "update"}, {"choices": sorted(GPT_names, key=custom_sort_key), "__type__": "update"}
+
+
+pretrained_sovits_name = "GPT_SoVITS/pretrained_models/s2G488k.pth"
+pretrained_gpt_name = "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"
+SoVITS_weight_root = "SoVITS_weights"
+GPT_weight_root = "GPT_weights"
+os.makedirs(SoVITS_weight_root, exist_ok=True)
+os.makedirs(GPT_weight_root, exist_ok=True)
+
+
+def get_weights_names():
+    SoVITS_names = [pretrained_sovits_name]
+    for name in os.listdir(SoVITS_weight_root):
+        if name.endswith(".pth"): SoVITS_names.append("%s/%s" % (SoVITS_weight_root, name))
+    GPT_names = [pretrained_gpt_name]
+    for name in os.listdir(GPT_weight_root):
+        if name.endswith(".ckpt"): GPT_names.append("%s/%s" % (GPT_weight_root, name))
+    return SoVITS_names, GPT_names
+
+
+SoVITS_names, GPT_names = get_weights_names()
 
 
 class GPTSoVITSGUI(QMainWindow):
+    gpt_path = gpt_path
+    sovits_path = sovits_path
+
     def __init__(self):
         super().__init__()
 
-        self.init_ui()
-
-    def init_ui(self):
         self.setWindowTitle('GPT-SoVITS GUI')
         self.setGeometry(800, 450, 950, 850)
 
@@ -71,6 +641,7 @@ class GPTSoVITSGUI(QMainWindow):
         self.GPT_model_label = QLabel("选择GPT模型:")
         self.GPT_model_input = QLineEdit()
         self.GPT_model_input.setPlaceholderText("拖拽或选择文件")
+        self.GPT_model_input.setText(self.gpt_path)
         self.GPT_model_input.setReadOnly(True)
         self.GPT_model_button = QPushButton("选择GPT模型文件")
         self.GPT_model_button.clicked.connect(self.select_GPT_model)
@@ -78,6 +649,7 @@ class GPTSoVITSGUI(QMainWindow):
         self.SoVITS_model_label = QLabel("选择SoVITS模型:")
         self.SoVITS_model_input = QLineEdit()
         self.SoVITS_model_input.setPlaceholderText("拖拽或选择文件")
+        self.SoVITS_model_input.setText(self.sovits_path)
         self.SoVITS_model_input.setReadOnly(True)
         self.SoVITS_model_button = QPushButton("选择SoVITS模型文件")
         self.SoVITS_model_button.clicked.connect(self.select_SoVITS_model)
@@ -91,25 +663,25 @@ class GPTSoVITSGUI(QMainWindow):
 
         self.ref_text_label = QLabel("参考音频文本:")
         self.ref_text_input = QLineEdit()
-        self.ref_text_input.setPlaceholderText("拖拽或选择文件")
+        self.ref_text_input.setPlaceholderText("直接输入文字或上传文本")
-        self.ref_text_input.setReadOnly(True)
         self.ref_text_button = QPushButton("上传文本")
         self.ref_text_button.clicked.connect(self.upload_ref_text)
 
-        self.language_label = QLabel("参考音频语言:")
+        self.ref_language_label = QLabel("参考音频语言:")
-        self.language_combobox = QComboBox()
+        self.ref_language_combobox = QComboBox()
-        self.language_combobox.addItems(["中文", "英文", "日文"])
+        self.ref_language_combobox.addItems(["中文", "英文", "日文", "中英混合", "日英混合", "多语种混合"])
+        self.ref_language_combobox.setCurrentText("多语种混合")
 
         self.target_text_label = QLabel("合成目标文本:")
         self.target_text_input = QLineEdit()
-        self.target_text_input.setPlaceholderText("拖拽或选择文件")
+        self.target_text_input.setPlaceholderText("直接输入文字或上传文本")
-        self.target_text_input.setReadOnly(True)
         self.target_text_button = QPushButton("上传文本")
         self.target_text_button.clicked.connect(self.upload_target_text)
 
-        self.language_label_02 = QLabel("合成音频语言:")
+        self.target_language_label = QLabel("合成音频语言:")
-        self.language_combobox_02 = QComboBox()
+        self.target_language_combobox = QComboBox()
-        self.language_combobox_02.addItems(["中文", "英文", "日文"])
+        self.target_language_combobox.addItems(["中文", "英文", "日文", "中英混合", "日英混合", "多语种混合"])
+        self.target_language_combobox.setCurrentText("多语种混合")
 
         self.output_label = QLabel("输出音频路径:")
         self.output_input = QLineEdit()
@@ -140,10 +712,8 @@ class GPTSoVITSGUI(QMainWindow):
 
         main_layout = QVBoxLayout()
 
-        input_layout = QGridLayout()
+        input_layout = QGridLayout(self)
-        input_layout.setSpacing(10) 
+        input_layout.setSpacing(10)
-
-        self.setLayout(input_layout)
 
         input_layout.addWidget(license_label, 0, 0, 1, 3)
 
@@ -159,22 +729,22 @@ class GPTSoVITSGUI(QMainWindow):
         input_layout.addWidget(self.ref_audio_input, 6, 0, 1, 2)
         input_layout.addWidget(self.ref_audio_button, 6, 2)
 
-        input_layout.addWidget(self.language_label, 7, 0)
+        input_layout.addWidget(self.ref_language_label, 7, 0)
-        input_layout.addWidget(self.language_combobox, 8, 0, 1, 1)
+        input_layout.addWidget(self.ref_language_combobox, 8, 0, 1, 1)
         input_layout.addWidget(self.ref_text_label, 9, 0)
         input_layout.addWidget(self.ref_text_input, 10, 0, 1, 2)
         input_layout.addWidget(self.ref_text_button, 10, 2)
 
-        input_layout.addWidget(self.language_label_02, 11, 0)
+        input_layout.addWidget(self.target_language_label, 11, 0)
-        input_layout.addWidget(self.language_combobox_02, 12, 0, 1, 1)
+        input_layout.addWidget(self.target_language_combobox, 12, 0, 1, 1)
         input_layout.addWidget(self.target_text_label, 13, 0)
         input_layout.addWidget(self.target_text_input, 14, 0, 1, 2)
         input_layout.addWidget(self.target_text_button, 14, 2)
-        
+
         input_layout.addWidget(self.output_label, 15, 0)
         input_layout.addWidget(self.output_input, 16, 0, 1, 2)
         input_layout.addWidget(self.output_button, 16, 2)
-        
+
         main_layout.addLayout(input_layout)
 
         output_layout = QVBoxLayout()
@@ -198,10 +768,8 @@ class GPTSoVITSGUI(QMainWindow):
     def dropEvent(self, event):
         if event.mimeData().hasUrls():
             file_paths = [url.toLocalFile() for url in event.mimeData().urls()]
-          
             if len(file_paths) == 1:
                 self.update_ref_audio(file_paths[0])
-                self.update_input_paths(self.ref_audio_input, file_paths[0])
             else:
                 self.update_ref_audio(", ".join(file_paths))
 
@@ -211,23 +779,13 @@ class GPTSoVITSGUI(QMainWindow):
             widget.installEventFilter(self)
 
     def eventFilter(self, obj, event):
-        if event.type() == QEvent.DragEnter:
+        if event.type() in (QEvent.DragEnter, QEvent.Drop):
             mime_data = event.mimeData()
             if mime_data.hasUrls():
                 event.acceptProposedAction()
-              
-        elif event.type() == QEvent.Drop:
-            mime_data = event.mimeData()
-            if mime_data.hasUrls():
-                file_paths = [url.toLocalFile() for url in mime_data.urls()]
-                if len(file_paths) == 1:
-                    self.update_input_paths(obj, file_paths[0])
-                else:
-                    self.update_input_paths(obj, ", ".join(file_paths))
-                event.acceptProposedAction()
 
         return super().eventFilter(obj, event)
-    
+
     def select_GPT_model(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择GPT模型文件", "", "GPT Files (*.ckpt)")
         if file_path:
@@ -239,24 +797,9 @@ class GPTSoVITSGUI(QMainWindow):
             self.SoVITS_model_input.setText(file_path)
 
     def select_ref_audio(self):
-        options = QFileDialog.Options()
+        file_path, _ = QFileDialog.getOpenFileName(self, "选择参考音频文件", "", "Audio Files (*.wav *.mp3)")
-        options |= QFileDialog.DontUseNativeDialog
+        if file_path:
-        options |= QFileDialog.ShowDirsOnly
+            self.update_ref_audio(file_path)
-
-        file_dialog = QFileDialog()
-        file_dialog.setOptions(options)
-
-        file_dialog.setFileMode(QFileDialog.AnyFile)
-        file_dialog.setNameFilter("Audio Files (*.wav *.mp3)")
-
-        if file_dialog.exec_():
-            file_paths = file_dialog.selectedFiles()
-          
-            if len(file_paths) == 1:
-                self.update_ref_audio(file_paths[0])
-                self.update_input_paths(self.ref_audio_input, file_paths[0])
-            else:
-                self.update_ref_audio(", ".join(file_paths))
 
     def upload_ref_text(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择文本文件", "", "Text Files (*.txt)")
@@ -264,7 +807,6 @@ class GPTSoVITSGUI(QMainWindow):
             with open(file_path, 'r', encoding='utf-8') as file:
                 content = file.read()
                 self.ref_text_input.setText(content)
-                self.update_input_paths(self.ref_text_input, file_path)
 
     def upload_target_text(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择文本文件", "", "Text Files (*.txt)")
@@ -272,7 +814,6 @@ class GPTSoVITSGUI(QMainWindow):
             with open(file_path, 'r', encoding='utf-8') as file:
                 content = file.read()
                 self.target_text_input.setText(content)
-                self.update_input_paths(self.target_text_input, file_path)
 
     def select_output_path(self):
         options = QFileDialog.Options()
@@ -290,9 +831,6 @@ class GPTSoVITSGUI(QMainWindow):
     def update_ref_audio(self, file_path):
         self.ref_audio_input.setText(file_path)
 
-    def update_input_paths(self, input_box, file_path):
-        input_box.setText(file_path)
-
     def clear_output(self):
         self.output_text.clear()
 
@@ -300,23 +838,27 @@ class GPTSoVITSGUI(QMainWindow):
         GPT_model_path = self.GPT_model_input.text()
         SoVITS_model_path = self.SoVITS_model_input.text()
         ref_audio_path = self.ref_audio_input.text()
-        language_combobox = self.language_combobox.currentText()
+        language_combobox = self.ref_language_combobox.currentText()
         language_combobox = i18n(language_combobox)
         ref_text = self.ref_text_input.text()
-        language_combobox_02 = self.language_combobox_02.currentText()
+        target_language_combobox = self.target_language_combobox.currentText()
-        language_combobox_02 = i18n(language_combobox_02)
+        target_language_combobox = i18n(target_language_combobox)
         target_text = self.target_text_input.text()
         output_path = self.output_input.text()
 
-        change_gpt_weights(gpt_path=GPT_model_path)
+        if GPT_model_path != self.gpt_path:
-        change_sovits_weights(sovits_path=SoVITS_model_path)
+            change_gpt_weights(gpt_path=GPT_model_path)
+            self.gpt_path = GPT_model_path
+        if SoVITS_model_path != self.sovits_path:
+            change_sovits_weights(sovits_path=SoVITS_model_path)
+            self.sovits_path = SoVITS_model_path
 
         synthesis_result = get_tts_wav(ref_wav_path=ref_audio_path, 
                                        prompt_text=ref_text, 
                                        prompt_language=language_combobox, 
                                        text=target_text, 
-                                       text_language=language_combobox_02)
+                                       text_language=target_language_combobox)
-        
+
         result_list = list(synthesis_result)
 
         if result_list:
@@ -329,12 +871,9 @@ class GPTSoVITSGUI(QMainWindow):
         self.status_bar.showMessage("合成完成！输出路径：" + output_wav_path, 5000)
         self.output_text.append("处理结果：\n" + result)
 
-def main():
+
+if __name__ == '__main__':
     app = QApplication(sys.argv)
     mainWin = GPTSoVITSGUI()
     mainWin.show()
-    sys.exit(app.exec_())
+    sys.exit(app.exec_())
-
-
-if __name__ == '__main__':
-    main()