Update app.py

app.py

@@ -4,11 +4,10 @@ import datetime
 from pathlib import Path
 import random
 
-# --- BAGIAN 1: DOWNLOAD OTOMATIS ASSET BERT ---
-# Menggunakan huggingface_hub untuk mengambil folder yang tidak kita upload
+# --- BAGIAN 1: DOWNLOAD OTOMATIS ASSET BERT (WAJIB AGAR JALAN) ---
 from huggingface_hub import snapshot_download
 
-# Mengaktifkan hf_transfer untuk kecepatan download maksimal (membutuhkan hf_transfer di requirements.txt)
+# Mengaktifkan hf_transfer untuk kecepatan download maksimal
 os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
 
 def download_bert_assets():
@@ -17,34 +16,31 @@ def download_bert_assets():
     SOURCE_SUBFOLDER = "sbv2-chupa-demo/bert"
     DEST_FOLDER = "./bert"
     
-    # Hanya download jika folder ./bert belum ada di Space
     if not os.path.exists(DEST_FOLDER):
         try:
             print(f"Downloading BERT assets from {REPO_ID}... Mohon tunggu sebentar.")
-            # Download folder bert saja dari subfolder repo sumber
             temp_dir = snapshot_download(
                 repo_id=REPO_ID,
                 allow_patterns=[f"{SOURCE_SUBFOLDER}/**/*"],
-                token=os.getenv("HF_TOKEN") # Mengambil token dari Secret Space
+                token=os.getenv("HF_TOKEN")
             )
             
-            # Pindahkan isi dari folder hasil download ke root/bert agar terbaca sistem
             src_path = os.path.join(temp_dir, SOURCE_SUBFOLDER)
             if os.path.exists(src_path):
                 shutil.copytree(src_path, DEST_FOLDER)
-                print("✅ BERT assets downloaded and linked successfully.")
+                print("✅ BERT assets downloaded successfully.")
             else:
-                print("⚠️ Folder bert tidak ditemukan di hasil download snapshot.")
+                print("⚠️ Folder bert tidak ditemukan.")
         except Exception as e:
             print(f"❌ Failed to download BERT: {e}")
     else:
         print("✅ BERT assets already exist.")
 
-# Jalankan download aset sebelum melakukan import library TTS
+# Jalankan download aset sebelum import library lainnya
 download_bert_assets()
 
 
-# --- BAGIAN 2: LOGIKA APLIKASI ASLI ---
+# --- BAGIAN 2: LOGIKA ASLI (TANPA UBAH TAMPILAN) ---
 import gradio as gr
 from style_bert_vits2.constants import (
     DEFAULT_LENGTH,
@@ -58,7 +54,7 @@ from style_bert_vits2.models.infer import InvalidToneError
 from style_bert_vits2.nlp.japanese import pyopenjtalk_worker as pyopenjtalk
 from style_bert_vits2.tts_model import TTSModelHolder
 
-# Inisialisasi worker untuk teks Jepang
+# Inisialisasi worker
 pyopenjtalk.initialize_worker()
 
 example_file = "chupa_examples.txt"
@@ -66,7 +62,7 @@ initial_text = (
     "ちゅぱ、ちゅるる、ぢゅ、んく、れーれゅれろれろれろ、じゅぽぽぽぽぽ……ちゅううう！"
 )
 
-# Load contoh teks jika file tersedia
+# Load examples
 if os.path.exists(example_file):
     with open(example_file, "r", encoding="utf-8") as f:
         examples = f.read().splitlines()
@@ -77,8 +73,9 @@ def get_random_text() -> str:
     return random.choice(examples)
 
 initial_md = """
-# チュパ音合成デモ (Chupa Sound Synthesis Demo)
-2024-07-07: initial ver - Migrated to Hugging Face Space
+# チュパ音合成デモ
+
+2024-07-07: initial ver
 """
 
 def make_interactive():
@@ -134,6 +131,7 @@ def create_inference_app(model_holder: TTSModelHolder) -> gr.Blocks:
 
         end_time = datetime.datetime.now()
         duration = (end_time - start_time).total_seconds()
+
         message = f"Success, time: {duration} seconds."
         return message, (sr, audio)
 
@@ -142,15 +140,16 @@ def create_inference_app(model_holder: TTSModelHolder) -> gr.Blocks:
 
     model_names = model_holder.model_names
     if len(model_names) == 0:
-        logger.error(f"モデルが見つかりませんでした。{model_holder.root_dir}にモデルを置いてください。")
+        logger.error(f"モデルが見つかりませんでした。")
         with gr.Blocks() as app:
-            gr.Markdown(f"Error: モデルが見つかりませんでした。{model_holder.root_dir}にモデルを置いてください。")
+            gr.Markdown(f"Error: モデルが見つかりませんでした。")
         return app
 
     initial_pth_files = get_model_files(model_names[0])
     model = model_holder.get_model(model_names[0], initial_pth_files[0])
     speakers = list(model.spk2id.keys())
 
+    # TAMPILAN ASLI (TIDAK DIUBAH)
     with gr.Blocks(theme="ParityError/Anime") as app:
         gr.Markdown(initial_md)
         with gr.Row():
@@ -177,24 +176,50 @@ def create_inference_app(model_holder: TTSModelHolder) -> gr.Blocks:
                     random_button.click(get_random_text, outputs=[text_input])
                 with gr.Row():
                     length_scale = gr.Slider(
-                        minimum=0.1, maximum=2, value=DEFAULT_LENGTH, step=0.1, label="生成音声の長さ（Length）"
+                        minimum=0.1,
+                        maximum=2,
+                        value=DEFAULT_LENGTH,
+                        step=0.1,
+                        label="生成音声の長さ（Length）",
                     )
                     sdp_ratio = gr.Slider(
-                        minimum=0, maximum=1, value=1, step=0.1, label="SDP Ratio"
+                        minimum=0,
+                        maximum=1,
+                        value=1,
+                        step=0.1,
+                        label="SDP Ratio",
                     )
                 line_split = gr.Checkbox(
-                    label="改行で分けて生成（分けたほうが感情が乗ります）",
+                    label="改行で分けて生成",
                     value=DEFAULT_LINE_SPLIT,
                     visible=False,
                 )
                 split_interval = gr.Slider(
-                    minimum=0.0, maximum=2, value=DEFAULT_SPLIT_INTERVAL, step=0.1, label="改行ごとに挟む無音の長さ（秒）"
+                    minimum=0.0,
+                    maximum=2,
+                    value=DEFAULT_SPLIT_INTERVAL,
+                    step=0.1,
+                    label="改行ごとに挟む無音の長さ（秒）",
+                )
+                language = gr.Dropdown(
+                    choices=["JP"], value="JP", label="Language", visible=False
                 )
-                language = gr.Dropdown(choices=["JP"], value="JP", label="Language", visible=False)
                 speaker = gr.Dropdown(label="話者", choices=speakers, value=speakers[0])
                 with gr.Accordion(label="詳細設定", open=True):
-                    noise_scale = gr.Slider(minimum=0.1, maximum=2, value=DEFAULT_NOISE, step=0.1, label="Noise")
-                    noise_scale_w = gr.Slider(minimum=0.1, maximum=2, value=DEFAULT_NOISEW, step=0.1, label="Noise_W")
+                    noise_scale = gr.Slider(
+                        minimum=0.1,
+                        maximum=2,
+                        value=DEFAULT_NOISE,
+                        step=0.1,
+                        label="Noise",
+                    )
+                    noise_scale_w = gr.Slider(
+                        minimum=0.1,
+                        maximum=2,
+                        value=DEFAULT_NOISEW,
+                        step=0.1,
+                        label="Noise_W",
+                    )
             with gr.Column():
                 tts_button = gr.Button("音声合成", variant="primary")
                 text_output = gr.Textbox(label="情報")
@@ -202,14 +227,24 @@ def create_inference_app(model_holder: TTSModelHolder) -> gr.Blocks:
 
         tts_button.click(
             tts_fn,
-            inputs=[model_name, model_path, text_input, language, sdp_ratio, noise_scale, noise_scale_w, length_scale, line_split, split_interval, speaker],
+            inputs=[
+                model_name, model_path, text_input, language, sdp_ratio,
+                noise_scale, noise_scale_w, length_scale, line_split,
+                split_interval, speaker
+            ],
             outputs=[text_output, audio_output],
         )
 
-        model_name.change(model_holder.update_model_files_for_gradio, inputs=[model_name], outputs=[model_path])
+        model_name.change(
+            model_holder.update_model_files_for_gradio,
+            inputs=[model_name],
+            outputs=[model_path],
+        )
         model_path.change(make_non_interactive, outputs=[tts_button])
-        refresh_button.click(model_holder.update_model_names_for_gradio, outputs=[model_name, model_path, tts_button])
-        
+        refresh_button.click(
+            model_holder.update_model_names_for_gradio,
+            outputs=[model_name, model_path, tts_button],
+        )
         style = gr.Dropdown(label="スタイル", choices=[], visible=False)
         load_button.click(
             model_holder.get_model_for_gradio,
@@ -219,17 +254,17 @@ def create_inference_app(model_holder: TTSModelHolder) -> gr.Blocks:
 
     return app
 
+
 if __name__ == "__main__":
     import torch
     from style_bert_vits2.constants import Languages
     from style_bert_vits2.nlp import bert_models
 
-    # Load model BERT yang sudah di-download sebelumnya
     bert_models.load_model(Languages.JP)
     bert_models.load_tokenizer(Languages.JP)
 
     device = "cuda" if torch.cuda.is_available() else "cpu"
-    # Pastikan folder model_assets sudah terisi file .safetensors/config
+    # Menggunakan folder model_assets yang sudah ada
     model_holder = TTSModelHolder(Path("model_assets"), device)
     app = create_inference_app(model_holder)
-    app.launch(inbrowser=True)
+    app.launch()