Update app.py

app.py

@@ -1,3 +1,4 @@
+import os
 import spaces
 import gradio as gr
 import torch
@@ -335,41 +336,46 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
             ).export(format="mp3", bitrate=bitrate).read()
             yield mp3_bytes, None
 
-
-if __name__ == "__main__":
-    description = ("State-of-the-Art zero-shot voice conversion/singing voice conversion. For local deployment please check [GitHub repository](https://github.com/Plachtaa/seed-vc) "
-                   "for details and updates.<br>Note that any reference audio will be forcefully clipped to 25s if beyond this length.<br> "
-                   "If total duration of source and reference audio exceeds 30s, source audio will be processed in chunks.<br> "
-                   "无需训练的 zero-shot 语音/歌声转换模型，若需本地部署查看[GitHub页面](https://github.com/Plachtaa/seed-vc)<br>"
-                   "请注意，参考音频若超过 25 秒，则会被自动裁剪至此长度。<br>若源音频和参考音频的总时长超过 30 秒，源音频将被分段处理。")
-    inputs = [
-        gr.Audio(type="filepath", label="Source Audio / 源音频"),
-        gr.Audio(type="filepath", label="Reference Audio / 参考音频"),
-        gr.Slider(minimum=1, maximum=200, value=25, step=1, label="Diffusion Steps / 扩散步数", info="25 by default, 50~100 for best quality / 默认为 25，50~100 为最佳质量"),
-        gr.Slider(minimum=0.5, maximum=2.0, step=0.1, value=1.0, label="Length Adjust / 长度调整", info="<1.0 for speed-up speech, >1.0 for slow-down speech / <1.0 加速语速，>1.0 减慢语速"),
-        gr.Slider(minimum=0.0, maximum=1.0, step=0.1, value=0.7, label="Inference CFG Rate", info="has subtle influence / 有微小影响"),
-        gr.Checkbox(label="Use F0 conditioned model / 启用F0输入", value=False, info="Must set to true for singing voice conversion / 歌声转换时必须勾选"),
-        gr.Checkbox(label="Auto F0 adjust / 自动F0调整", value=True,
-                    info="Roughly adjust F0 to match target voice. Only works when F0 conditioned model is used. / 粗略调整 F0 以匹配目标音色，仅在勾选 '启用F0输入' 时生效"),
-        gr.Slider(label='Pitch shift / 音调变换', minimum=-24, maximum=24, step=1, value=0, info="Pitch shift in semitones, only works when F0 conditioned model is used / 半音数的音高变换，仅在勾选 '启用F0输入' 时生效"),
-    ]
-
-    examples = [["examples/source/yae_0.wav", "examples/reference/dingzhen_0.wav", 25, 1.0, 0.7, False, True, 0],
-                ["examples/source/jay_0.wav", "examples/reference/azuma_0.wav", 25, 1.0, 0.7, False, True, 0],
-                ["examples/source/Wiz Khalifa,Charlie Puth - See You Again [vocals]_[cut_28sec].wav",
-                 "examples/reference/kobe_0.wav", 50, 1.0, 0.7, True, False, -6],
-                ["examples/source/TECHNOPOLIS - 2085 [vocals]_[cut_14sec].wav",
-                 "examples/reference/trump_0.wav", 50, 1.0, 0.7, True, False, -12],
-                ]
-
-    outputs = [gr.Audio(label="Stream Output Audio / 流式输出", streaming=True, format='mp3'),
-               gr.Audio(label="Full Output Audio / 完整输出", streaming=False, format='wav')]
-
-    gr.Interface(fn=voice_conversion,
-                 description=description,
-                 inputs=inputs,
-                 outputs=outputs,
-                 title="Seed Voice Conversion",
-                 examples=examples,
-                 cache_examples=False,
-                 ).launch()
+default_dir = "/default"
+reference_files = [
+    ("ずんだもん", "zundamon"),
+    ("四国めたん", "sikokumetan"),
+    ("春日部つむぎ", "kasukabetsumugi"),
+    ("雨晴はう", "ameharehau"),
+    ("波音リツ", "namineritsu"),
+]
+
+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}.png"), name, filename] for name, filename in reference_files]
+
+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。"
+               "また、元音声と参考音声の合計時間が30秒を超える場合、元音声は分割処理されます。")
+
+inputs = [
+    gr.Audio(type="filepath", label="元音声"),
+    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"),
+    gr.Slider(minimum=0.0, maximum=1.0, step=0.1, value=0.7, label="推論CFG率", info="わずかな影響があります"),
+    gr.Checkbox(label="F0条件付きモデルを使用", value=False, info="歌声変換には必須です"),
+    gr.Checkbox(label="F0自動調整", value=True, info="F0をおおよそ調整して目標音声に合わせます。F0条件付きモデル使用時にのみ有効です"),
+    gr.Slider(label='音程変換', minimum=-24, maximum=24, step=1, value=0, info="半音単位の音程変換。F0条件付きモデル使用時にのみ有効です"),
+]
+
+outputs = [
+    gr.Audio(label="ストリーム出力音声", streaming=True, format='mp3'),
+    gr.Audio(label="完全出力音声", streaming=False, format='wav')
+]
+
+gr.Interface(
+    fn=proxy_voice_conversion,
+    description=description,
+    inputs=inputs,
+    outputs=outputs,
+    title="Seed Voice Conversion with Reference Gallery",
+    cache_examples=False,
+).launch()