Refactor: dataset_root and assets_root in configs/paths.yml

.gitignore

@@ -17,3 +17,5 @@ venv/
 
 /scripts/test/
 *.zip
+*.csv
+*.bak

app.py

@@ -4,9 +4,8 @@ import os
 import sys
 
 import gradio as gr
-import numpy as np
 import torch
-from gradio.processing_utils import convert_to_16_bit_wav
+import yaml
 
 from common.constants import (
     DEFAULT_ASSIST_TEXT_WEIGHT,
@@ -22,7 +21,12 @@ from common.constants import (
 )
 from common.log import logger
 from common.tts_model import ModelHolder
-from config import config
+
+# Get path settings
+with open(os.path.join("configs", "paths.yml"), "r", encoding="utf-8") as f:
+    path_config: dict[str, str] = yaml.safe_load(f.read())
+    # dataset_root = path_config["dataset_root"]
+    assets_root = path_config["assets_root"]
 
 languages = [l.value for l in Languages]
 
@@ -182,7 +186,7 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--cpu", action="store_true", help="Use CPU instead of GPU")
     parser.add_argument(
-        "--dir", "-d", type=str, help="Model directory", default=config.out_dir
+        "--dir", "-d", type=str, help="Model directory", default=assets_root
     )
     parser.add_argument(
         "--share", action="store_true", help="Share this app publicly", default=False

colab.ipynb

@@ -6,9 +6,10 @@
       "source": [
         "# Style-Bert-VITS2 (ver 1.2) のGoogle Colabでの学習\n",
         "\n",
-        "Google driveと連携することで、Google Colab上でStyle-Bert-VITS2の学習を行うことができます。\n",
+        "Google Colab上でStyle-Bert-VITS2の学習を行うことができます。\n",
         "\n",
-        "このnotebookでは、あなたのGoogle Driveにフォルダ`Style-Bert-VITS2`を作り、その内部での作業を行います。他のフォルダには触れません。\n",
+        "このnotebookでは、通常使用ではあなたのGoogle Driveにフォルダ`Style-Bert-VITS2`を作り、その内部での作業を行います。他のフォルダには触れません。\n",
+        "Google Driveを使わない場合は、初期設定のところで適切なパスを指定してください。\n",
         "\n",
         "## 流れ\n",
         "\n",
@@ -16,16 +17,75 @@
         "上から順に実行していけばいいです。音声合成に必要なファイルはGoogle Driveの`Style-Bert-VITS2/model_assets/`に保存されます。また、途中経過も`Style-Bert-VITS2/Data/`に保存されるので、学習を中断したり、途中から再開することもできます。\n",
         "\n",
         "### 学習を途中から再開したいとき\n",
-        "1と2を行い、3の前処理は飛ばして、4から始めてください。\n"
+        "0と1と2を行い、3の前処理は飛ばして、4から始めてください。\n"
       ]
     },
     {
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "## 1. Google Driveとの連携とデータの配置\n",
+        "## 0. 環境構築\n",
         "\n",
-        "次のセルを実行して、Google driveと連携します。"
+        "Style-Bert-VITS2の環境をcolab上に構築します。グラボモードが有効になっていることを確認し、以下のセルを順に実行してください。"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "#@title このセルを実行して環境構築してください。\n",
+        "#@markdown 最後に赤文字でエラーや警告が出ても何故かうまくいくみたいです。\n",
+        "\n",
+        "!git clone https://github.com/litagin02/Style-Bert-VITS2.git\n",
+        "%cd Style-Bert-VITS2/\n",
+        "!pip install -r requirements.txt"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "#@title 必要なモデルのダウンロード\n",
+        "\n",
+        "!python initialize.py"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 1. 初期設定\n",
+        "\n",
+        "学習とその結果を保存するディレクトリ名を指定します。\n",
+        "Google driveの場合はそのまま実行、カスタマイズしたい方は変更して実行してください。"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 1,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "dataset_root = \"/content/drive/MyDrive/Style-Bert-VITS2/Data\"\n",
+        "assets_root = \"/content/drive/MyDrive/Style-Bert-VITS2/model_assets\"\n",
+        "\n",
+        "import yaml\n",
+        "\n",
+        "with open(\"configs/paths.yml\", \"w\", encoding=\"utf-8\") as f:\n",
+        "    yaml.dump({\"dataset_root\": dataset_root, \"assets_root\": assets_root}, f)"
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## 2. Google Driveとの連携とデータの配置\n",
+        "\n",
+        "Google driveにデータを保存する方は、次のセルを実行して、Google driveと連携します。"
       ]
     },
     {
@@ -48,7 +108,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "次のセルを実行して、Google driveの中にStyle-Bert-VITS2フォルダと、学習データをいれるDataフォルダを作成します。"
+        "次のセルを実行して、学習データをいれるフォルダ（1で設定した`dataset_root`）を作成します。"
       ]
     },
     {
@@ -61,7 +121,7 @@
       "source": [
         "import os\n",
         "\n",
-        "os.makedirs((\"/content/drive/MyDrive/Style-Bert-VITS2/Data/\"), exist_ok=True)"
+        "os.makedirs(dataset_root, exist_ok=True)"
       ]
     },
     {
@@ -104,55 +164,6 @@
         "```"
       ]
     },
-    {
-      "cell_type": "markdown",
-      "metadata": {},
-      "source": [
-        "## 2. 環境構築\n",
-        "\n",
-        "Style-Bert-VITS2の環境をcolab上に構築します。グラボモードが有効になっていることを確認し、以下のセルを順に実行してください。"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "cellView": "form",
-        "colab": {
-          "base_uri": "https://localhost:8080/",
-          "height": 1000
-        },
-        "id": "0CH9aI_MOmTA",
-        "outputId": "34cf1e8f-23b9-4d2b-e040-ea5a3ec44a82"
-      },
-      "outputs": [],
-      "source": [
-        "#@title このセルを実行して環境構築してください。\n",
-        "#@markdown 最後に赤文字でエラーや警告が出ても何故かうまくいくみたいです。\n",
-        "\n",
-        "!git clone https://github.com/litagin02/Style-Bert-VITS2.git\n",
-        "%cd Style-Bert-VITS2/\n",
-        "!pip install -r requirements.txt"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "cellView": "form",
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "CpLs5YoNPe4Z",
-        "outputId": "0edd3309-08d1-4372-e47a-6a805a74b620"
-      },
-      "outputs": [],
-      "source": [
-        "#@title 必要なモデルのダウンロード\n",
-        "\n",
-        "!python initialize.py"
-      ]
-    },
     {
       "cell_type": "markdown",
       "metadata": {
@@ -178,7 +189,8 @@
         "# 学習のバッチサイズ。4ぐらいが最適か。VRAMのはみ出具合に応じて調整してください。\n",
         "batch_size = 4\n",
         "\n",
-        "# 学習のエポック数（データセットを合計何周するか）。100ぐらいで十分だと思われます。\n",
+        "# 学習のエポック数（データセットを合計何周するか）。\n",
+        "# 100ぐらいで十分かもしれませんが、もっと多くやると質が上がるのかもしれません。\n",
         "epochs = 100\n",
         "\n",
         "# 保存頻度。何ステップごとにモデルを保存するか。分からなければデフォルトのままで。\n",
@@ -256,15 +268,15 @@
         "from webui_train import get_path\n",
         "\n",
         "dataset_path, _, _, _, config_path = get_path(model_name)\n",
+        "\n",
         "with open(\"default_config.yml\", \"r\", encoding=\"utf-8\") as f:\n",
         "    yml_data = yaml.safe_load(f)\n",
         "yml_data[\"model_name\"] = model_name\n",
-        "yml_data[\"dataset_path\"] = dataset_path\n",
         "with open(\"config.yml\", \"w\", encoding=\"utf-8\") as f:\n",
         "    yaml.dump(yml_data, f, allow_unicode=True)\n",
         "\n",
         "\n",
-        "!python train_ms.py --config {config_path} --model {dataset_path}"
+        "!python train_ms.py --config {config_path} --model {dataset_path} --asset_root {assets_root}"
       ]
     },
     {
@@ -280,7 +292,7 @@
       "outputs": [],
       "source": [
         "#@title 学習結果を試すならここから\n",
-        "!python app.py --share --dir \"/content/drive/MyDrive/Style-Bert-VITS2/model_assets/\""
+        "!python app.py --share --dir {assets_root}"
       ]
     }
   ],
@@ -295,7 +307,16 @@
       "name": "python3"
     },
     "language_info": {
-      "name": "python"
+      "codemirror_mode": {
+        "name": "ipython",
+        "version": 3
+      },
+      "file_extension": ".py",
+      "mimetype": "text/x-python",
+      "name": "python",
+      "nbconvert_exporter": "python",
+      "pygments_lexer": "ipython3",
+      "version": "3.10.11"
     }
   },
   "nbformat": 4,

config.py

@@ -205,7 +205,7 @@ class Translate_config:
 
 
 class Config:
-    def __init__(self, config_path: str):
+    def __init__(self, config_path: str, path_config: dict[str, str]):
         if not os.path.isfile(config_path) and os.path.isfile("default_config.yml"):
             shutil.copy(src="default_config.yml", dst=config_path)
             print(
@@ -222,12 +222,10 @@ class Config:
             if "dataset_path" in yaml_config:
                 dataset_path = yaml_config["dataset_path"]
             else:
-                dataset_path = f"Data/{model_name}"
+                dataset_path = os.path.join(path_config["dataset_root"], model_name)
-            self.out_dir = yaml_config["out_dir"]
-            # openi_token: str = yaml_config["openi_token"]
             self.dataset_path: str = dataset_path
-            # self.mirror: str = yaml_config["mirror"]
+            self.assets_root: str = path_config["assets_root"]
-            # self.openi_token: str = openi_token
+            self.out_dir = os.path.join(self.assets_root, model_name)
             self.resample_config: Resample_config = Resample_config.from_dict(
                 dataset_path, yaml_config["resample"]
             )
@@ -256,14 +254,27 @@ class Config:
             # )
 
 
+with open(os.path.join("configs", "paths.yml"), "r", encoding="utf-8") as f:
+    path_config: dict[str, str] = yaml.safe_load(f.read())
+    # Should contain the following keys:
+    # - dataset_root: the root directory of the dataset, default to "Data"
+    # - assets_root: the root directory of the assets, default to "model_assets"
+
 parser = argparse.ArgumentParser()
 # 为避免与以前的config.json起冲突，将其更名如下
-parser.add_argument("-y", "--yml_config", type=str, default="config.yml")
+parser.add_argument(
-args, _ = parser.parse_known_args()
+    "-y",
+    "--yml_config",
+    type=str,
+    default="config.yml",
+    help="Path to setting yaml file.",
+)
+
+args = parser.parse_args()
 
 try:
-    config = Config(args.yml_config)
+    config = Config(args.yml_config, path_config)
 except TypeError:
     logger.warning("Old config.yml found. Replace it with default_config.yml.")
     shutil.copy(src="default_config.yml", dst="config.yml")
-    config = Config("config.yml")
+    config = Config("config.yml", path_config)

configs/paths.yml

@@ -0,0 +1,8 @@
+# Root directory of the training dataset.
+# The training dataset of {model_name} should be placed in {dataset_root}/{model_name}.
+dataset_root: Data
+
+# Root directory of the model assets (for inference).
+# In training, the model assets will be saved to {assets_root}/{model_name},
+# and in inference, we load all the models from {assets_root}.
+assets_root: model_assets

default_config.yml

@@ -1,24 +1,20 @@
-# Global configuration file for Bert-VITS2
-
 model_name: "model_name"
 
-out_dir: "model_assets"
-
 # If you want to use a specific dataset path, uncomment the following line.
-# Otherwise, the dataset path is `Data/{model_name}`.
+# Otherwise, the dataset path is `{dataset_root}/{model_name}`.
 
 # dataset_path: "your/dataset/path"
 
 resample:
   sampling_rate: 44100
-  in_dir: "audios/raw"
+  in_dir: "raw"
-  out_dir: "audios/wavs"
+  out_dir: "wavs"
 
 preprocess_text:
-  transcription_path: "filelists/esd.list"
+  transcription_path: "esd.list"
   cleaned_path: ""
-  train_path: "filelists/train.list"
+  train_path: "train.list"
-  val_path: "filelists/val.list"
+  val_path: "val.list"
   config_path: "config.json"
   val_per_lang: 4
   max_val_total: 12
@@ -42,13 +38,13 @@ train_ms:
     WORLD_SIZE: 1
     LOCAL_RANK: 0
     RANK: 0
-  model: "models"
+  model_dir: "models"  # The directory to save the model (for training), relative to `{dataset_root}/{model_name}`.
   config_path: "config.json"
   num_workers: 16
   spec_cache: True
   keep_ckpts: 1 # Set this to 0 to keep all checkpoints
 
-webui:
+webui:  # For `webui.py`, which is not supported yet in Style-Bert-VITS2.
   # 推理设备
   device: "cuda"
   # 模型路径

server_fastapi.py

@@ -67,7 +67,7 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--cpu", action="store_true", help="Use CPU instead of GPU")
     parser.add_argument(
-        "--dir", "-d", type=str, help="Model directory", default=config.out_dir
+        "--dir", "-d", type=str, help="Model directory", default=config.assets_root
     )
     args = parser.parse_args()
 

train_ms.py

@@ -19,6 +19,8 @@ from tqdm import tqdm
 import commons
 import default_style
 import utils
+from common.log import logger
+from common.stdout_wrapper import SAFE_STDOUT
 from config import config
 from data_utils import (
     DistributedBucketSampler,
@@ -29,8 +31,6 @@ from losses import discriminator_loss, feature_loss, generator_loss, kl_loss
 from mel_processing import mel_spectrogram_torch, spec_to_mel_torch
 from models import DurationDiscriminator, MultiPeriodDiscriminator, SynthesizerTrn
 from text.symbols import symbols
-from common.log import logger
-from common.stdout_wrapper import SAFE_STDOUT
 
 torch.backends.cuda.matmul.allow_tf32 = True
 torch.backends.cudnn.allow_tf32 = (
@@ -44,12 +44,6 @@ torch.backends.cuda.enable_mem_efficient_sdp(
 )  # Not available if torch version is lower than 2.0
 torch.backends.cuda.enable_math_sdp(True)
 
-try:
-    import google.colab
-
-    IS_COLAB = True
-except ImportError:
-    IS_COLAB = False
 
 global_step = 0
 
@@ -86,7 +80,7 @@ def run():
     # 命令行/config.yml配置解析
     # hps = utils.get_hparams()
     parser = argparse.ArgumentParser()
-    # 非必要不建议使用命令行配置，请使用config.yml文件
+    # Command line configuration is not recommended unless necessary, use config.yml
     parser.add_argument(
         "-c",
         "--config",
@@ -102,12 +96,26 @@ def run():
         help="数据集文件夹路径，请注意，数据不再默认放在/logs文件夹下。如果需要用命令行配置，请声明相对于根目录的路径",
         default=config.dataset_path,
     )
+    parser.add_argument(
+        "--assets_root",
+        type=str,
+        help="Root directory of model assets needed for inference.",
+        default=config.assets_root,
+    )
+    parser.add_argument(
+        "--skip_default_style",
+        action="store_true",
+        help="Skip saving default style config and mean vector.",
+    )
     args = parser.parse_args()
     model_dir = os.path.join(args.model, config.train_ms_config.model)
     if not os.path.exists(model_dir):
         os.makedirs(model_dir)
     hps = utils.get_hparams_from_file(args.config)
+
+    # This is needed because we have to pass `model_dir` to `train_and_evaluate()`
     hps.model_dir = model_dir
+
     # 比较路径是否相同
     if os.path.realpath(args.config) != os.path.realpath(
         config.train_ms_config.config_path
@@ -125,30 +133,26 @@ def run():
 
     args.model: For saving all info needed for training.
         default: `Data/{model_name}`.
-    hps.model_dir = model_dir: For saving checkpoints (for resuming training).
+    hps.model_dir := model_dir: For saving checkpoints (for resuming training).
         default: `Data/{model_name}/models`.
+        (Use `hps` since we have to pass `model_dir` to `train_and_evaluate()`.
 
-    config.out_dir: Root directory of model assets needed for inference.
+    args.assets_root: The root directory of model assets needed for inference.
-        default: `model_assets`.
+        default: config.assets_root == `model_assets`.
 
-    out_dir: For saving resulting models (for inference).
+    config.out_dir: The directory for model assets of this model (for inference).
-        default: `model_assets/{model_name}`, which is used for inference.
+        default: `model_assets/{model_name}`.
     """
-    if IS_COLAB:
-        config.out_dir = "/content/drive/MyDrive/Style-Bert-VITS2/model_assets"
-        logger.info(
-            "Colab detected, so use mounted Google Drive as directory for saving resulting models:"
-        )
-        logger.info(config.out_dir)
     os.makedirs(config.out_dir, exist_ok=True)
-    out_dir = os.path.join(config.out_dir, config.model_name)
-    os.makedirs(out_dir, exist_ok=True)
 
+    if not args.skip_default_style:
         # Save default style to out_dir
-    default_style.set_style_config(args.config, os.path.join(out_dir, "config.json"))
+        default_style.set_style_config(
+            args.config, os.path.join(config.out_dir, "config.json")
+        )
         default_style.save_mean_vector(
             os.path.join(args.model, "wavs"),
-        os.path.join(out_dir, "style_vectors.npy"),
+            os.path.join(config.out_dir, "style_vectors.npy"),
         )
 
     torch.manual_seed(hps.train.seed)
@@ -158,10 +162,10 @@ def run():
     if rank == 0:
         # logger = utils.get_logger(hps.model_dir)
         # logger.info(hps)
-        logger.add(os.path.join(hps.model_dir, "train.log"))
+        logger.add(os.path.join(model_dir, "train.log"))
-        utils.check_git_hash(hps.model_dir)
+        utils.check_git_hash(model_dir)
-        writer = SummaryWriter(log_dir=hps.model_dir)
+        writer = SummaryWriter(log_dir=model_dir)
-        writer_eval = SummaryWriter(log_dir=os.path.join(hps.model_dir, "eval"))
+        writer_eval = SummaryWriter(log_dir=os.path.join(model_dir, "eval"))
     train_dataset = TextAudioSpeakerLoader(hps.data.training_files, hps.data)
     train_sampler = DistributedBucketSampler(
         train_dataset,
@@ -284,10 +288,10 @@ def run():
             net_dur_disc, device_ids=[local_rank], find_unused_parameters=True
         )
 
-    if utils.is_resuming(hps.model_dir):
+    if utils.is_resuming(model_dir):
         if net_dur_disc is not None:
             _, _, dur_resume_lr, epoch_str = utils.load_checkpoint(
-                utils.latest_checkpoint_path(hps.model_dir, "DUR_*.pth"),
+                utils.latest_checkpoint_path(model_dir, "DUR_*.pth"),
                 net_dur_disc,
                 optim_dur_disc,
                 skip_optimizer=hps.train.skip_optimizer
@@ -297,7 +301,7 @@ def run():
             if not optim_dur_disc.param_groups[0].get("initial_lr"):
                 optim_dur_disc.param_groups[0]["initial_lr"] = dur_resume_lr
         _, optim_g, g_resume_lr, epoch_str = utils.load_checkpoint(
-            utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"),
+            utils.latest_checkpoint_path(model_dir, "G_*.pth"),
             net_g,
             optim_g,
             skip_optimizer=hps.train.skip_optimizer
@@ -305,7 +309,7 @@ def run():
             else True,
         )
         _, optim_d, d_resume_lr, epoch_str = utils.load_checkpoint(
-            utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"),
+            utils.latest_checkpoint_path(model_dir, "D_*.pth"),
             net_d,
             optim_d,
             skip_optimizer=hps.train.skip_optimizer
@@ -320,7 +324,7 @@ def run():
         epoch_str = max(epoch_str, 1)
         # global_step = (epoch_str - 1) * len(train_loader)
         global_step = int(
-            utils.get_steps(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"))
+            utils.get_steps(utils.latest_checkpoint_path(model_dir, "G_*.pth"))
         )
         logger.info(
             f"******************检测到模型存在，epoch为 {epoch_str}，gloabl step为 {global_step}*********************"
@@ -328,14 +332,14 @@ def run():
     else:
         try:
             _ = utils.load_safetensors(
-                os.path.join(hps.model_dir, "G_0.safetensors"), net_g
+                os.path.join(model_dir, "G_0.safetensors"), net_g
             )
             _ = utils.load_safetensors(
-                os.path.join(hps.model_dir, "D_0.safetensors"), net_d
+                os.path.join(model_dir, "D_0.safetensors"), net_d
             )
             if net_dur_disc is not None:
                 _ = utils.load_safetensors(
-                    os.path.join(hps.model_dir, "DUR_0.safetensors"), net_dur_disc
+                    os.path.join(model_dir, "DUR_0.safetensors"), net_dur_disc
                 )
             logger.info("Loaded the pretrained models.")
         except Exception as e:
@@ -402,14 +406,14 @@ def run():
                 optim_g,
                 hps.train.learning_rate,
                 epoch,
-                os.path.join(hps.model_dir, "G_{}.pth".format(global_step)),
+                os.path.join(model_dir, "G_{}.pth".format(global_step)),
             )
             utils.save_checkpoint(
                 net_d,
                 optim_d,
                 hps.train.learning_rate,
                 epoch,
-                os.path.join(hps.model_dir, "D_{}.pth".format(global_step)),
+                os.path.join(model_dir, "D_{}.pth".format(global_step)),
             )
             if net_dur_disc is not None:
                 utils.save_checkpoint(
@@ -417,13 +421,13 @@ def run():
                     optim_dur_disc,
                     hps.train.learning_rate,
                     epoch,
-                    os.path.join(hps.model_dir, "DUR_{}.pth".format(global_step)),
+                    os.path.join(model_dir, "DUR_{}.pth".format(global_step)),
                 )
             utils.save_safetensors(
                 net_g,
                 epoch,
                 os.path.join(
-                    out_dir,
+                    config.out_dir,
                     f"{config.model_name}_e{epoch}_s{global_step}.safetensors",
                 ),
                 for_infer=True,
@@ -690,12 +694,12 @@ def train_and_evaluate(
                         n_ckpts_to_keep=keep_ckpts,
                         sort_by_time=True,
                     )
+                # Save safetensors (for inference) to `model_assets/{model_name}`
                 utils.save_safetensors(
                     net_g,
                     epoch,
                     os.path.join(
                         config.out_dir,
-                        config.model_name,
                         f"{config.model_name}_e{epoch}_s{global_step}.safetensors",
                     ),
                     for_infer=True,

webui_dataset.py

@@ -1,15 +1,22 @@
 import os
 
 import gradio as gr
+import yaml
 
 from common.log import logger
-from common.subprocess_utils import run_script_with_log, second_elem_of
+from common.subprocess_utils import run_script_with_log
+
+# Get path settings
+with open(os.path.join("configs", "paths.yml"), "r", encoding="utf-8") as f:
+    path_config: dict[str, str] = yaml.safe_load(f.read())
+    dataset_root = path_config["dataset_root"]
+    # assets_root = path_config["assets_root"]
 
 
 def do_slice(model_name, normalize):
     logger.info("Start slicing...")
     input_dir = "inputs"
-    output_dir = os.path.join("Data", model_name, "raw")
+    output_dir = os.path.join(dataset_root, model_name, "raw")
     cmd = [
         "slice.py",
         "--input_dir",
@@ -26,8 +33,8 @@ def do_slice(model_name, normalize):
 
 
 def do_transcribe(model_name):
-    input_dir = os.path.join("Data", model_name, "raw")
+    input_dir = os.path.join(dataset_root, model_name, "raw")
-    output_file = os.path.join("Data", model_name, "esd.list")
+    output_file = os.path.join(dataset_root, model_name, "esd.list")
     result = run_script_with_log(
         [
             "transcribe.py",

webui_merge.py

@@ -5,14 +5,13 @@ import sys
 import gradio as gr
 import numpy as np
 import torch
+import yaml
 from safetensors import safe_open
 from safetensors.torch import save_file
 
-
-from config import config
-from common.tts_model import Model, ModelHolder
-from common.log import logger
 from common.constants import DEFAULT_STYLE
+from common.log import logger
+from common.tts_model import Model, ModelHolder
 
 voice_keys = ["dec", "flow"]
 speech_style_keys = ["enc_p"]
@@ -20,8 +19,13 @@ tempo_keys = ["sdp", "dp"]
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-model_dir = config.out_dir
+# Get path settings
-model_holder = ModelHolder(model_dir, device)
+with open(os.path.join("configs", "paths.yml"), "r", encoding="utf-8") as f:
+    path_config: dict[str, str] = yaml.safe_load(f.read())
+    # dataset_root = path_config["dataset_root"]
+    assets_root = path_config["assets_root"]
+
+model_holder = ModelHolder(assets_root, device)
 
 
 def merge_style(model_name_a, model_name_b, weight, output_name, style_triple_list):

webui_style_vectors.py

@@ -115,7 +115,7 @@ def do_clustering_gradio(n_clusters=4, method="KMeans"):
 
 def save_style_vectors(model_name, style_names: str):
     """centerとcentroidsを保存する"""
-    result_dir = os.path.join(config.out_dir, model_name)
+    result_dir = os.path.join(config.assets_root, model_name)
     os.makedirs(result_dir, exist_ok=True)
     style_vectors = np.stack([mean] + centroids)
     style_vector_path = os.path.join(result_dir, "style_vectors.npy")
@@ -148,7 +148,7 @@ def save_style_vectors_from_files(model_name, audio_files_text, style_names_text
         return "Error: スタイルベクトルを読み込んでください。"
     mean = np.mean(x, axis=0)
 
-    result_dir = os.path.join(config.out_dir, model_name)
+    result_dir = os.path.join(config.assets_root, model_name)
     os.makedirs(result_dir, exist_ok=True)
     audio_files = audio_files_text.split(",")
     style_names = style_names_text.split(",")

webui_train.py

@@ -1,7 +1,6 @@
 import json
 import os
 import shutil
-import sys
 from multiprocessing import cpu_count
 
 import gradio as gr
@@ -10,24 +9,16 @@ import yaml
 from common.log import logger
 from common.subprocess_utils import run_script_with_log, second_elem_of
 
-try:
+# Get path settings
-    import google.colab
+with open(os.path.join("configs", "paths.yml"), "r", encoding="utf-8") as f:
-
+    path_config: dict[str, str] = yaml.safe_load(f.read())
-    IS_COLAB = True
+    dataset_root = path_config["dataset_root"]
-except ImportError:
+    # assets_root = path_config["assets_root"]
-    IS_COLAB = False
 
 
 def get_path(model_name):
     assert model_name != "", "モデル名は空にできません"
-    if IS_COLAB:
+    dataset_path = os.path.join(dataset_root, model_name)
-        logger.info("Colab detected, so use mounted Google Drive as dataset path:")
-        dataset_path = os.path.join(
-            "/content/drive/MyDrive/Style-Bert-VITS2/Data", model_name
-        )
-        logger.info(dataset_path)
-    else:
-        dataset_path = os.path.join("Data", model_name)
     lbl_path = os.path.join(dataset_path, "esd.list")
     train_path = os.path.join(dataset_path, "train.list")
     val_path = os.path.join(dataset_path, "val.list")
@@ -61,7 +52,6 @@ def initialize(model_name, batch_size, epochs, save_every_steps, bf16_run):
         )
     except FileExistsError:
         logger.warning(f"Step 1: {model_path} already exists.")
-        return False, f"Step1, Error: モデルフォルダ {model_path} が既に存在します。問題なければ削除してください。"
     except FileNotFoundError:
         logger.error("Step 1: `pretrained` folder not found.")
         return False, "Step 1, Error: pretrainedフォルダが見つかりません。"