Update app.py

app.py

@@ -341,11 +341,10 @@ reference_files = [
     ("四国めたん", "sikokumetan")
 ]
 
-def proxy_voice_conversion(source, selected_reference, diffusion_steps, length_adjust, inference_cfg_rate, f0_condition, auto_f0_adjust, pitch_shift):
-    reference_path = os.path.join(default_dir, f"{selected_reference}.mp3")
-    return voice_conversion(source, reference_path, diffusion_steps, length_adjust, inference_cfg_rate, f0_condition, auto_f0_adjust, pitch_shift)
 
-gallery_items = [[os.path.join(default_dir, f"{filename}.webp"), name] for name, filename in reference_files]
+gallery_items = [[os.path.join(default_dir, f"{filename}.webp"), name, filename] for name, filename in reference_files]
+
+default_reference = os.path.join(default_dir, f"{reference_files[0][1]}.mp3")
 
 description = ("Zero-shot音声変換モデル（学習不要）。ローカルでの利用方法は[GitHubリポジトリ](https://github.com/Plachtaa/seed-vc)をご覧ください。"
                "参考音声が25秒を超える場合、自動的に25秒にクリップされます。"
@@ -353,6 +352,7 @@ description = ("Zero-shot音声変換モデル（学習不要）。ローカル
 
 inputs = [
     gr.Audio(type="filepath", label="元音声"),
+    gr.Audio(type="filepath", label="参考音声", visible=False, value=default_reference),
     gr.Gallery(label="参照音声を選択", value=gallery_items, columns=5, interactive=True),
     gr.Slider(minimum=1, maximum=200, value=10, step=1, label="拡散ステップ数", info="デフォルトは10、50～100が最適な品質"),
     gr.Slider(minimum=0.5, maximum=2.0, step=0.1, value=1.0, label="長さ調整", info="1.0未満で速度を上げ、1.0以上で速度を遅くします"),
@@ -368,10 +368,10 @@ outputs = [
 ]
 
 gr.Interface(
-    fn=proxy_voice_conversion,
+    fn=voice_conversion,
     description=description,
     inputs=inputs,
     outputs=outputs,
     title="Seed Voice Conversion with Reference Gallery",
     cache_examples=False,
-).launch()
+).launch()