添加说话人特征注入以及相关处理

.gitignore

@@ -10,3 +10,4 @@ logs
 reference
 SoVITS_weights
 GPT_weights
+TEMP

GPT_SoVITS/AR/data/data_module.py

@@ -1,7 +1,7 @@
 # modified from https://github.com/feng-yufei/shared_debugging_code/blob/main/data_module.py
 from pytorch_lightning import LightningDataModule
 from AR.data.bucket_sampler import DistributedBucketSampler
-from AR.data.dataset import Text2SemanticDataset
+from AR.data.dataset import Text2SemanticDataset, Text2SemanticSpeakerDataset
 from torch.utils.data import DataLoader
 
 
@@ -26,12 +26,20 @@ class Text2SemanticDataModule(LightningDataModule):
         pass
 
     def setup(self, stage=None, output_logs=False):
-        self._train_dataset = Text2SemanticDataset(
+        if 'train_dir' not in self.config:
-            phoneme_path=self.train_phoneme_path,
+            self._train_dataset = Text2SemanticDataset(
-            semantic_path=self.train_semantic_path,
+                phoneme_path=self.train_phoneme_path,
-            max_sec=self.config["data"]["max_sec"],
+                semantic_path=self.train_semantic_path,
-            pad_val=self.config["data"]["pad_val"],
+                max_sec=self.config["data"]["max_sec"],
-        )
+                pad_val=self.config["data"]["pad_val"],
+            )
+        else:
+            self._train_dataset = Text2SemanticSpeakerDataset(
+                self.config['speaker_dict_path'],
+                self.config['train_dir'],
+                max_sec=self.config["data"]["max_sec"],
+                pad_val=self.config["data"]["pad_val"],
+            )
         self._dev_dataset = self._train_dataset
         # self._dev_dataset = Text2SemanticDataset(
         #     phoneme_path=self.dev_phoneme_path,

GPT_SoVITS/AR/data/dataset.py

@@ -293,6 +293,128 @@ class Text2SemanticDataset(Dataset):
             "bert_feature": bert_padded,
         }
 
+class Text2SemanticSpeakerDataset(Dataset):
+    def __init__(
+        self,
+        speaker_dict_path: str,
+        train_dir: str,
+        max_sample: int = None,
+        max_sec: int = 100,
+        pad_val: int = 1024,
+        # min value of phoneme/sec
+        min_ps_ratio: int = 3,
+        # max value of phoneme/sec
+        max_ps_ratio: int = 25,
+    ):
+        super().__init__()
+        self.speaker_dict = dict()
+        if os.path.exists(speaker_dict_path):
+            with open(speaker_dict_path, "r", encoding="utf-8") as f:
+                try:
+                    self.speaker_dict = json.load(f)
+                except:
+                    pass
+        
+        self.speaker2dataset = {}
+        self.PAD = pad_val
+        for name in os.listdir(train_dir):
+            if name in ["logs_s1", os.path.basename(speaker_dict_path)]:
+                continue
+            opt_dir = os.path.join(train_dir, name)
+            speaker_id = self.speaker_dict.get(name, None)
+            if speaker_id is None:
+                speaker_id = len(self.speaker_dict)
+                self.speaker_dict[name] = speaker_id
+            self.speaker2dataset[speaker_id] = Text2SemanticDataset(
+                phoneme_path=os.path.join(opt_dir, "2-name2text.txt"),
+                semantic_path=os.path.join(opt_dir, "6-name2semantic.tsv"),
+                max_sample=max_sample,
+                max_sec=max_sec,
+                pad_val=pad_val,
+                min_ps_ratio=min_ps_ratio,
+                max_ps_ratio=max_ps_ratio,
+            )
+        with open(speaker_dict_path, "w", encoding="utf-8") as f:
+            json.dump(self.speaker_dict, f, ensure_ascii=False, indent=4)
+    
+    def __len__(self):
+        result = 0
+        for _, dataset in self.speaker2dataset.items():
+            result += len(dataset)
+        return result
+    
+    def __getitem__(self, idx):
+        for speaker_id, dataset in self.speaker2dataset.items():
+            if idx < len(dataset):
+                result = dataset[idx]
+                result["speaker_id"] = speaker_id
+                return result
+            else:
+                idx -= len(dataset)
+        raise IndexError("index out of range")
+
+    def get_sample_length(self, idx: int):
+        for speaker_id, dataset in self.speaker2dataset.items():
+            if idx < len(dataset):
+                semantic_ids = dataset.semantic_phoneme[idx][0]
+                sec = 1.0 * len(semantic_ids) / dataset.hz
+                return sec
+            else:
+                idx -= len(dataset)
+        raise IndexError("index out of range")
+
+    def collate(self, examples: List[Dict]) -> Dict:
+        sample_index: List[int] = []
+        phoneme_ids: List[torch.Tensor] = []
+        phoneme_ids_lens: List[int] = []
+        semantic_ids: List[torch.Tensor] = []
+        semantic_ids_lens: List[int] = []
+        speaker_ids: List[int] = []
+        # return
+
+        for item in examples:
+            sample_index.append(item["idx"])
+            phoneme_ids.append(np.array(item["phoneme_ids"], dtype=np.int64))
+            semantic_ids.append(np.array(item["semantic_ids"], dtype=np.int64))
+            phoneme_ids_lens.append(item["phoneme_ids_len"])
+            semantic_ids_lens.append(item["semantic_ids_len"])
+            speaker_ids.append(item["speaker_id"])
+
+        # pad 0
+        phoneme_ids = batch_sequences(phoneme_ids)
+        semantic_ids = batch_sequences(semantic_ids, pad_value=self.PAD)
+
+        # # convert each batch to torch.tensor
+        phoneme_ids = torch.tensor(phoneme_ids)
+        semantic_ids = torch.tensor(semantic_ids)
+        phoneme_ids_lens = torch.tensor(phoneme_ids_lens)
+        semantic_ids_lens = torch.tensor(semantic_ids_lens)
+        bert_padded = torch.FloatTensor(len(examples), 1024, max(phoneme_ids_lens))
+        bert_padded.zero_()
+        speaker_ids = torch.tensor(speaker_ids, dtype=torch.long)
+
+        for idx, item in enumerate(examples):
+            bert = item["bert_feature"]
+            if bert != None:
+                bert_padded[idx, :, : bert.shape[-1]] = bert
+
+        return {
+            # List[int]
+            "ids": sample_index,
+            # torch.Tensor (B, max_phoneme_length)
+            "phoneme_ids": phoneme_ids,
+            # torch.Tensor (B)
+            "phoneme_ids_len": phoneme_ids_lens,
+            # torch.Tensor (B, max_semantic_ids_length)
+            "semantic_ids": semantic_ids,
+            # torch.Tensor (B)
+            "semantic_ids_len": semantic_ids_lens,
+            # torch.Tensor (B, 1024, max_phoneme_length)
+            "bert_feature": bert_padded,
+            "speaker_ids": speaker_ids,
+        }
+        
+        
 
 if __name__ == "__main__":
     root_dir = "/data/docker/liujing04/gpt-vits/prepare/dump_mix/"

GPT_SoVITS/AR/models/t2s_lightning_module.py

@@ -34,10 +34,11 @@ class Text2SemanticLightningModule(LightningModule):
             self.eval_dir.mkdir(parents=True, exist_ok=True)
         for param in self.model.parameters():
             param.requires_grad = False
-        self.model.speaker_proj.weight.requires_grad = True
+        self.model.speaker_proj.requires_grad_(True)
-        self.model.speaker_proj.bias.requires_grad = True
+        self.model.speaker_proj.weight.requires_grad_(True)
-        self.model.speaker_proj.train()
+        self.model.speaker_proj.bias.requires_grad_(True)
-        self.model.speaker_feat.requires_grad = True
+        self.model.speaker_emb.requires_grad_(True)
+        self.model.speaker_emb.weight.requires_grad_(True)
 
     def training_step(self, batch: Dict, batch_idx: int):
         opt = self.optimizers()
@@ -48,13 +49,13 @@ class Text2SemanticLightningModule(LightningModule):
             batch["semantic_ids"],
             batch["semantic_ids_len"],
             batch["bert_feature"],
+            speaker_ids=batch["speaker_ids"] if "speaker_ids" in batch else None,
         )
         self.manual_backward(loss)
         if batch_idx > 0 and batch_idx % 4 == 0:
             opt.step()
             opt.zero_grad()
             scheduler.step()
-            torch.save(self.model.speaker_feat.data, "C:/Users/86150/Desktop/GPT-SoVITS/zyj.pt")
 
         self.log(
             "total_loss",

GPT_SoVITS/AR/models/t2s_model.py

@@ -52,12 +52,8 @@ class Text2SemanticDecoder(nn.Module):
         # should be same as num of kmeans bin
         # assert self.EOS == 1024
         self.bert_proj = nn.Linear(1024, self.embedding_dim)
-        self.speaker_proj = nn.Linear(1024, self.embedding_dim)
+        self.speaker_proj = nn.Linear(4096, self.embedding_dim)
-        self.path_speaker = "C:/Users/86150/Desktop/GPT-SoVITS/zyj.pt"
+        self.speaker_emb = nn.Embedding(100, 4096)
-        if not os.path.exists(self.path_speaker):
-            self.speaker_feat = nn.Parameter(torch.randn(1024) * 0.1)
-        else:
-            self.speaker_feat = nn.Parameter(torch.load(self.path_speaker, map_location="cpu"))
         self.ar_text_embedding = TokenEmbedding(
             self.embedding_dim, self.phoneme_vocab_size, self.p_dropout
         )
@@ -95,7 +91,7 @@ class Text2SemanticDecoder(nn.Module):
             ignore_index=self.EOS,
         )
 
-    def make_input_data(self, x, x_lens, y, y_lens, bert_feature):
+    def make_input_data(self, x, x_lens, y, y_lens, bert_feature, speaker_ids=None):
         x = self.ar_text_embedding(x)
         x = x + self.bert_proj(bert_feature.transpose(1, 2))
         x = self.ar_text_position(x)
@@ -111,7 +107,8 @@ class Text2SemanticDecoder(nn.Module):
         x_len = x_lens.max()
         y_len = y_lens.max()
         y_emb = self.ar_audio_embedding(y)
-        y_emb = y_emb + self.speaker_proj(self.speaker_feat).view(1,1,-1)
+        if speaker_ids is not None:
+            y_emb = y_emb + self.speaker_proj(self.speaker_emb(speaker_ids)).unsqueeze(1)
         y_pos = self.ar_audio_position(y_emb)
 
         xy_padding_mask = torch.concat([x_mask, y_mask], dim=1)
@@ -149,7 +146,7 @@ class Text2SemanticDecoder(nn.Module):
 
         return xy_pos, xy_attn_mask, targets
 
-    def forward(self, x, x_lens, y, y_lens, bert_feature):
+    def forward(self, x, x_lens, y, y_lens, bert_feature, speaker_ids=None):
         """
         x: phoneme_ids
         y: semantic_ids
@@ -157,7 +154,7 @@ class Text2SemanticDecoder(nn.Module):
 
         reject_y, reject_y_lens = make_reject_y(y, y_lens)
 
-        xy_pos, xy_attn_mask, targets = self.make_input_data(x, x_lens, y, y_lens, bert_feature)
+        xy_pos, xy_attn_mask, targets = self.make_input_data(x, x_lens, y, y_lens, bert_feature, speaker_ids=speaker_ids)
 
         xy_dec, _ = self.h(
             (xy_pos, None),
@@ -167,7 +164,7 @@ class Text2SemanticDecoder(nn.Module):
         logits = self.ar_predict_layer(xy_dec[:, x_len:])
 
         ###### DPO #############
-        reject_xy_pos, reject_xy_attn_mask, reject_targets = self.make_input_data(x, x_lens, reject_y, reject_y_lens, bert_feature)
+        reject_xy_pos, reject_xy_attn_mask, reject_targets = self.make_input_data(x, x_lens, reject_y, reject_y_lens, bert_feature, speaker_ids=speaker_ids)
 
         reject_xy_dec, _ = self.h(
             (reject_xy_pos, None),
@@ -338,7 +335,7 @@ class Text2SemanticDecoder(nn.Module):
         top_p: int = 100,
         early_stop_num: int = -1,
         temperature: float = 1.0,
-        use_speaker_feat=True,
+        speaker_ids=None,
     ):
         x = self.ar_text_embedding(x)
         x = x + self.bert_proj(bert_feature.transpose(1, 2))
@@ -362,8 +359,8 @@ class Text2SemanticDecoder(nn.Module):
             "first_infer": 1,
             "stage": 0,
         }
-        if use_speaker_feat:
+        if speaker_ids is not None:
-            speaker_feat = self.speaker_proj(self.speaker_feat).view(1,1,-1)
+            speaker_feat = self.speaker_proj(self.speaker_emb(speaker_ids)).unsqueeze(1)
         else:
             speaker_feat = 0
         for idx in tqdm(range(1500)):

GPT_SoVITS/configs/s1longer.yaml

@@ -3,7 +3,7 @@ train:
   epochs: 20
   batch_size: 8
   save_every_n_epoch: 1
-  precision: 16-mixed
+  precision: "32"
   gradient_clip: 1.0
 optimizer:
   lr: 0.01

GPT_SoVITS/inference_webui.py

@@ -365,7 +365,10 @@ def merge_short_text_in_array(texts, threshold):
             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):
+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,speaker_id="-1"):
+    speaker_id = int(speaker_id)
+    if speaker_id == -1:
+        speaker_id = None
     t0 = ttime()
     prompt_language = dict_language[prompt_language]
     text_language = dict_language[text_language]
@@ -381,7 +384,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
     )
     with torch.no_grad():
         wav16k, sr = librosa.load(ref_wav_path, sr=16000)
-        if (wav16k.shape[0] > 160000 or wav16k.shape[0] < 48000):
+        if (wav16k.shape[0] > 160000 or wav16k.shape[0] < 0):
             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)
@@ -448,6 +451,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
                 top_p=top_p,
                 temperature=temperature,
                 early_stop_num=hz * max_sec,
+                speaker_ids=torch.LongTensor([speaker_id]).to(device) if speaker_id is not None else None,
             )
         t3 = ttime()
         # print(pred_semantic.shape,idx)
@@ -601,6 +605,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             GPT_dropdown = gr.Dropdown(label=i18n("GPT模型列表"), choices=sorted(GPT_names, key=custom_sort_key), value=gpt_path, interactive=True)
             SoVITS_dropdown = gr.Dropdown(label=i18n("SoVITS模型列表"), choices=sorted(SoVITS_names, key=custom_sort_key), value=sovits_path, interactive=True)
             refresh_button = gr.Button(i18n("刷新模型路径"), variant="primary")
+            speaker_id = gr.Dropdown(label="说话人idx", choices=[str(i) for i in range(-1, 100)], value="-1")
             refresh_button.click(fn=change_choices, inputs=[], outputs=[SoVITS_dropdown, GPT_dropdown])
             SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown], [])
             GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], [])
@@ -632,7 +637,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
 
         inference_button.click(
             get_tts_wav,
-            [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut,top_k,top_p,temperature],
+            [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut,top_k,top_p,temperature,speaker_id],
             [output],
         )
 

GPT_SoVITS/prepare_datasets/2-get-hubert-wav32k.py

@@ -98,7 +98,7 @@ for line in lines[int(i_part)::int(all_parts)]:
     try:
         # wav_name,text=line.split("\t")
         wav_name, spk_name, language, text = line.split("|")
-        if (inp_wav_dir !=None):
+        if (inp_wav_dir !=None and len(inp_wav_dir) > 0):
             wav_name = os.path.basename(wav_name)
             wav_path = "%s/%s"%(inp_wav_dir, wav_name)
 

tools/slice_audio.py

@@ -24,24 +24,36 @@ def slice(inp,opt_root,threshold,min_length,min_interval,hop_size,max_sil_kept,_
     )
     _max=float(_max)
     alpha=float(alpha)
-    for inp_path in input[int(i_part)::int(all_part)]:
+    
-        # print(inp_path)
+    def slice_audio(root_dir, inp_wav_path):
         try:
-            name = os.path.basename(inp_path)
+            name = os.path.basename(inp_wav_path)
-            audio = load_audio(inp_path, 32000)
+            audio = load_audio(inp_wav_path, 32000)
             # print(audio.shape)
             for chunk, start, end in slicer.slice(audio):  # start和end是帧数
                 tmp_max = np.abs(chunk).max()
                 if(tmp_max>1):chunk/=tmp_max
                 chunk = (chunk / tmp_max * (_max * alpha)) + (1 - alpha) * chunk
                 wavfile.write(
-                    "%s/%s_%010d_%010d.wav" % (opt_root, name, start, end),
+                    "%s/%s_%010d_%010d.wav" % (root_dir, name, start, end),
                     32000,
                     # chunk.astype(np.float32),
                     (chunk * 32767).astype(np.int16),
                 )
         except:
-            print(inp_path,"->fail->",traceback.format_exc())
+            print(inp_wav_path,"->fail->",traceback.format_exc())
+    
+    for inp_path in input[int(i_part)::int(all_part)]:
+        # print(inp_path)
+        if os.path.isdir(inp_path):
+            inp_dir_name = os.path.basename(inp_path)
+            os.makedirs(os.path.join(opt_root, inp_dir_name), exist_ok=True)
+            for inp_name in os.listdir(inp_path):
+                inp_wav_path = os.path.join(inp_path, inp_name)
+                slice_audio(os.path.join(opt_root, inp_dir_name), inp_wav_path)
+        else:
+            inp_wav_path = inp_path
+            slice_audio(opt_root, inp_wav_path)
     return "执行完毕，请检查输出文件"
 
 print(slice(*sys.argv[1:]))

webui.py

@@ -195,20 +195,42 @@ def change_tts_inference(if_tts,bert_path,cnhubert_base_path,gpu_number,gpt_path
 from tools.asr.config import asr_dict
 def open_asr(asr_inp_dir, asr_opt_dir, asr_model, asr_model_size, asr_lang):
     global p_asr
-    if(p_asr==None):
+    def run(asr_inp_dir, asr_opt_dir):
-        asr_inp_dir=my_utils.clean_path(asr_inp_dir)
         cmd = f'"{python_exec}" tools/asr/{asr_dict[asr_model]["path"]}'
         cmd += f' -i "{asr_inp_dir}"'
         cmd += f' -o "{asr_opt_dir}"'
         cmd += f' -s {asr_model_size}'
         cmd += f' -l {asr_lang}'
         cmd += " -p %s"%("float16"if is_half==True else "float32")
-
-        yield "ASR任务开启：%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
         print(cmd)
         p_asr = Popen(cmd, shell=True)
         p_asr.wait()
         p_asr=None
+    if(p_asr==None):
+        # yield "ASR任务开启：%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
+        flag = False
+        for name in os.listdir(asr_inp_dir):
+            if os.path.isdir(os.path.join(asr_inp_dir, name)):
+                os.makedirs(os.path.join(asr_opt_dir, name), exist_ok=True)
+                run(os.path.join(asr_inp_dir, name), os.path.join(asr_opt_dir, name))
+            else:
+                break
+        else:
+            flag = True
+        if flag:
+            run(asr_inp_dir, asr_opt_dir)
+        # asr_inp_dir=my_utils.clean_path(asr_inp_dir)
+        # cmd = f'"{python_exec}" tools/asr/{asr_dict[asr_model]["path"]}'
+        # cmd += f' -i "{asr_inp_dir}"'
+        # cmd += f' -o "{asr_opt_dir}"'
+        # cmd += f' -s {asr_model_size}'
+        # cmd += f' -l {asr_lang}'
+        # cmd += " -p %s"%("float16"if is_half==True else "float32")
+
+        # print(cmd)
+        # p_asr = Popen(cmd, shell=True)
+        # p_asr.wait()
+        # p_asr=None
         yield f"ASR任务完成, 查看终端进行下一步",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
     else:
         yield "已有正在进行的ASR任务，需先终止才能开启下一次任务",{"__type__":"update","visible":False},{"__type__":"update","visible":True}
@@ -273,7 +295,7 @@ def open1Bb(batch_size,total_epoch,exp_name,if_save_latest,if_save_every_weights
             data=f.read()
             data=yaml.load(data, Loader=yaml.FullLoader)
         s1_dir="%s/%s"%(exp_root,exp_name)
-        os.makedirs("%s/logs_s1"%(s1_dir),exist_ok=True)
+        # os.makedirs("%s/logs_s1"%(s1_dir),exist_ok=True)
         if(is_half==False):
             data["train"]["precision"]="32"
             batch_size = max(1, batch_size // 2)
@@ -287,6 +309,8 @@ def open1Bb(batch_size,total_epoch,exp_name,if_save_latest,if_save_every_weights
         data["train"]["exp_name"]=exp_name
         data["train_semantic_path"]="%s/6-name2semantic.tsv"%s1_dir
         data["train_phoneme_path"]="%s/2-name2text.txt"%s1_dir
+        data["train_dir"] = s1_dir
+        data["speaker_dict_path"] = "%s/speaker_dict.json"%s1_dir
         data["output_dir"]="%s/logs_s1"%s1_dir
 
         os.environ["_CUDA_VISIBLE_DEVICES"]=gpu_numbers.replace("-",",")
@@ -352,6 +376,13 @@ def close_slice():
 ps1a=[]
 def open1a(inp_text,inp_wav_dir,exp_name,gpu_numbers,bert_pretrained_dir):
     global ps1a
+    if not inp_text.endswith(".list"):
+        for name in os.listdir(inp_text):
+            p = os.path.join(inp_text, name)
+            l = os.path.join(p, name + ".list")
+            assert os.path.exists(l)
+            yield from open1a(l,inp_wav_dir,os.path.join(exp_name, name),gpu_numbers,bert_pretrained_dir)
+        return
     inp_text = my_utils.clean_path(inp_text)
     inp_wav_dir = my_utils.clean_path(inp_wav_dir)
     if (ps1a == []):
@@ -410,6 +441,13 @@ def close1a():
 ps1b=[]
 def open1b(inp_text,inp_wav_dir,exp_name,gpu_numbers,ssl_pretrained_dir):
     global ps1b
+    if not inp_text.endswith(".list"):
+        for name in os.listdir(inp_text):
+            p = os.path.join(inp_text, name)
+            l = os.path.join(p, name + ".list")
+            assert os.path.exists(l)
+            yield from open1b(l,inp_wav_dir,os.path.join(exp_name, name),gpu_numbers,ssl_pretrained_dir)
+        return
     inp_text = my_utils.clean_path(inp_text)
     inp_wav_dir = my_utils.clean_path(inp_wav_dir)
     if (ps1b == []):
@@ -458,6 +496,13 @@ def close1b():
 ps1c=[]
 def open1c(inp_text,exp_name,gpu_numbers,pretrained_s2G_path):
     global ps1c
+    if not inp_text.endswith(".list"):
+        for name in os.listdir(inp_text):
+            p = os.path.join(inp_text, name)
+            l = os.path.join(p, name + ".list")
+            assert os.path.exists(l)
+            yield from open1c(l,os.path.join(exp_name, name),gpu_numbers,pretrained_s2G_path)
+        return
     inp_text = my_utils.clean_path(inp_text)
     if (ps1c == []):
         opt_dir="%s/%s"%(exp_root,exp_name)
@@ -515,6 +560,13 @@ def close1c():
 ps1abc=[]
 def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numbers1c,bert_pretrained_dir,ssl_pretrained_dir,pretrained_s2G_path):
     global ps1abc
+    if not inp_text.endswith(".list"):
+        for name in os.listdir(inp_text):
+            p = os.path.join(inp_text, name)
+            l = os.path.join(p, name + ".list")
+            assert os.path.exists(l)
+            yield from open1abc(l,inp_wav_dir,os.path.join(exp_name, name),gpu_numbers1a,gpu_numbers1Ba,gpu_numbers1c,bert_pretrained_dir,ssl_pretrained_dir,pretrained_s2G_path)
+        return
     inp_text = my_utils.clean_path(inp_text)
     inp_wav_dir = my_utils.clean_path(inp_wav_dir)
     if (ps1abc == []):