Update app.py

app.py

@@ -4,11 +4,18 @@ import shutil
 import tempfile
 from fastapi import FastAPI, UploadFile, File, Form, HTTPException
 from fastapi.responses import FileResponse
-from core.cloner import KokoClone
 
-# ── Chargement unique des modèles ─────────────────────────────────────────────
-print("Loading KokoClone models...")
-cloner = KokoClone()
+# ── Singleton : évite le double chargement ────────────────────────────────────
+_cloner = None
+
+def get_cloner():
+    global _cloner
+    if _cloner is None:
+        from core.cloner import KokoClone
+        print("Loading KokoClone models (once)...")
+        _cloner = KokoClone()
+        print("KokoClone ready.")
+    return _cloner
 
 
 # ── Handlers Gradio ───────────────────────────────────────────────────────────
@@ -18,7 +25,7 @@ def gradio_speak(text, lang, speed):
         raise gr.Error("Veuillez entrer du texte.")
     output_file = "/tmp/gradio_speak_output.wav"
     try:
-        cloner.speak(text=text, lang=lang, output_path=output_file, speed=float(speed))
+        get_cloner().speak(text=text, lang=lang, output_path=output_file, speed=float(speed))
         return output_file
     except Exception as e:
         raise gr.Error(f"Erreur : {str(e)}")
@@ -31,7 +38,7 @@ def gradio_clone(text, lang, ref_audio_path):
         raise gr.Error("Please upload or record a reference audio file.")
     output_file = "/tmp/gradio_clone_output.wav"
     try:
-        cloner.generate(text=text, lang=lang, reference_audio=ref_audio_path, output_path=output_file)
+        get_cloner().generate(text=text, lang=lang, reference_audio=ref_audio_path, output_path=output_file)
         return output_file
     except Exception as e:
         raise gr.Error(f"An error occurred: {str(e)}")
@@ -44,15 +51,15 @@ def gradio_convert(source_audio_path, ref_audio_path):
         raise gr.Error("Please upload a reference audio file.")
     output_file = "/tmp/gradio_convert_output.wav"
     try:
-        cloner.convert(source_audio=source_audio_path, reference_audio=ref_audio_path, output_path=output_file)
+        get_cloner().convert(source_audio=source_audio_path, reference_audio=ref_audio_path, output_path=output_file)
         return output_file
     except Exception as e:
         raise gr.Error(f"An error occurred: {str(e)}")
 
 
-# ── Interface Gradio (Gradio 6 compatible) ────────────────────────────────────
+# ── Interface Gradio ──────────────────────────────────────────────────────────
 
-with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
+with gr.Blocks() as demo:
     gr.Markdown(
         """
         <div style="text-align: center;">
@@ -64,11 +71,8 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
 
     with gr.Tabs():
 
-        # ── Onglet 1 : TTS Simple ─────────────────────────────────────────────
         with gr.Tab("🗣️ Simple TTS"):
-            gr.Markdown(
-                "Génère la parole avec une voix Kokoro — **aucun audio de référence requis**."
-            )
+            gr.Markdown("Génère la parole avec une voix Kokoro — **aucun audio de référence requis**.")
             with gr.Row():
                 with gr.Column(scale=1):
                     speak_text = gr.Textbox(
@@ -96,13 +100,8 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
                         step=0.1
                     )
                     speak_btn = gr.Button("🗣️ Générer l'audio", variant="primary")
-
                 with gr.Column(scale=1):
-                    speak_out = gr.Audio(
-                        label="Audio généré",
-                        interactive=False,
-                        autoplay=True
-                    )
+                    speak_out = gr.Audio(label="Audio généré", interactive=False, autoplay=True)
                     gr.Markdown(
                         """
 **Voix disponibles**
@@ -115,49 +114,30 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
 > Vitesse **0.7–0.8** recommandée pour les apprenants débutants.
                         """
                     )
-
             speak_btn.click(
                 fn=gradio_speak,
                 inputs=[speak_text, speak_lang, speak_speed],
                 outputs=speak_out
             )
 
-        # ── Onglet 2 : Text → Clone ───────────────────────────────────────────
         with gr.Tab("🎤 Text → Clone"):
             gr.Markdown("Synthétise le texte puis applique la voix d'un audio de référence.")
             with gr.Row():
                 with gr.Column(scale=1):
-                    clone_text = gr.Textbox(
-                        label="1. Texte à synthétiser",
-                        lines=4,
-                        placeholder="Enter the text you want spoken..."
-                    )
+                    clone_text = gr.Textbox(label="1. Texte à synthétiser", lines=4)
                     clone_lang = gr.Dropdown(
                         label="2. Langue",
                         choices=[
-                            ("English", "en"),
-                            ("Français", "fr"),
-                            ("Hindi", "hi"),
-                            ("Japanese", "ja"),
-                            ("Chinese", "zh"),
-                            ("Español", "es"),
-                            ("Italiano", "it"),
-                            ("Português", "pt"),
+                            ("English", "en"), ("Français", "fr"), ("Hindi", "hi"),
+                            ("Japanese", "ja"), ("Chinese", "zh"), ("Español", "es"),
+                            ("Italiano", "it"), ("Português", "pt"),
                         ],
                         value="en"
                     )
-                    clone_ref = gr.Audio(
-                        label="3. Audio de référence",
-                        type="filepath"
-                    )
+                    clone_ref = gr.Audio(label="3. Audio de référence", type="filepath")
                     clone_btn = gr.Button("🚀 Générer avec clonage", variant="primary")
-
                 with gr.Column(scale=1):
-                    clone_out = gr.Audio(
-                        label="Audio cloné",
-                        interactive=False,
-                        autoplay=False
-                    )
+                    clone_out = gr.Audio(label="Audio cloné", interactive=False, autoplay=False)
                     gr.Markdown(
                         """
 **Conseils**
@@ -166,34 +146,21 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
 - La langue doit correspondre au texte
                         """
                     )
-
             clone_btn.click(
                 fn=gradio_clone,
                 inputs=[clone_text, clone_lang, clone_ref],
                 outputs=clone_out
             )
 
-        # ── Onglet 3 : Audio → Clone ──────────────────────────────────────────
         with gr.Tab("🔁 Audio → Clone"):
             gr.Markdown("Re-voice un audio existant pour qu'il sonne comme la voix de référence.")
             with gr.Row():
                 with gr.Column(scale=1):
-                    convert_src = gr.Audio(
-                        label="1. Audio source",
-                        type="filepath"
-                    )
-                    convert_ref = gr.Audio(
-                        label="2. Voix de référence",
-                        type="filepath"
-                    )
+                    convert_src = gr.Audio(label="1. Audio source", type="filepath")
+                    convert_ref = gr.Audio(label="2. Voix de référence", type="filepath")
                     convert_btn = gr.Button("🔁 Convertir la voix", variant="primary")
-
                 with gr.Column(scale=1):
-                    convert_out = gr.Audio(
-                        label="Audio converti",
-                        interactive=False,
-                        autoplay=False
-                    )
+                    convert_out = gr.Audio(label="Audio converti", interactive=False, autoplay=False)
                     gr.Markdown(
                         """
 **Comment ça marche**
@@ -203,7 +170,6 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
 - Aucune transcription nécessaire
                         """
                     )
-
             convert_btn.click(
                 fn=gradio_convert,
                 inputs=[convert_src, convert_ref],
@@ -211,7 +177,7 @@ with gr.Blocks() as demo:  # theme retiré ici → va dans launch()
             )
 
 
-# ── FastAPI montée sur Gradio ─────────────────────────────────────────────────
+# ── FastAPI + Gradio ──────────────────────────────────────────────────────────
 
 app = gr.mount_gradio_app(FastAPI(), demo, path="/")
 
@@ -238,7 +204,7 @@ async def api_speak(
 ):
     out_path = tempfile.mktemp(suffix=".wav")
     try:
-        cloner.speak(text=text, lang=lang, output_path=out_path, speed=speed)
+        get_cloner().speak(text=text, lang=lang, output_path=out_path, speed=speed)
         return FileResponse(out_path, media_type="audio/wav", filename="speech.wav")
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
@@ -257,7 +223,7 @@ async def api_clone(
         ref_path = ref_tmp.name
     out_path = ref_path.replace(".wav", "_out.wav")
     try:
-        cloner.generate(text=text, lang=lang, reference_audio=ref_path, output_path=out_path)
+        get_cloner().generate(text=text, lang=lang, reference_audio=ref_path, output_path=out_path)
         return FileResponse(out_path, media_type="audio/wav", filename="cloned.wav")
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
@@ -279,7 +245,7 @@ async def api_convert(
         ref_path = ref_tmp.name
     out_path = src_path.replace(".wav", "_converted.wav")
     try:
-        cloner.convert(source_audio=src_path, reference_audio=ref_path, output_path=out_path)
+        get_cloner().convert(source_audio=src_path, reference_audio=ref_path, output_path=out_path)
         return FileResponse(out_path, media_type="audio/wav", filename="converted.wav")
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))