Update app.py

app.py

@@ -5,7 +5,7 @@ from huggingface_hub import snapshot_download
 from nemo.collections import asr as nemo_asr
 import gradio as gr
 
-# 1. CONFIGURATION
+# 1. CONFIGURATION MATÉRIEL ET LISTE DES MODÈLES ROBOTSMALI
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
 MODELS = {
@@ -23,65 +23,98 @@ MODELS = {
     "Traduction Soloni (ST)":    ("RobotsMali/st-soloni-114m-tdt-ctc", "rnnt"),
 }
 
-# Recherche de la vidéo d'exemple dans le dossier courant
-EXAMPLE_VIDEO = "MARALINKE.mp4" if os.path.exists("MARALINKE.mp4") else None
+# --- OPTIMISATION : DETECTION DE LA VIDEO DANS LE DOSSIER EXAMPLES ---
+def find_example_video():
+    # Liste des noms possibles basés sur ta capture d'écran
+    paths = [
+        "examples/MARALINKE_FIXED.mp4", 
+        "examples/MARALINKE.mp4",
+        "MARALINKE.mp4"
+    ]
+    for p in paths:
+        if os.path.exists(p):
+            return p
+    return None
+
+EXAMPLE_PATH = find_example_video()
 
 _cache = {}
 
+# 2. GESTION DE LA MÉMOIRE ET CHARGEMENT DU MODÈLE
+def clear_memory():
+    _cache.clear()
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+
 def get_model(name):
     if name in _cache: return _cache[name]
-    gc.collect()
-    if torch.cuda.is_available(): torch.cuda.empty_cache()
-    
+    clear_memory()
     repo, _ = MODELS[name]
+    
+    print(f"📥 Téléchargement de {repo}...")
     folder = snapshot_download(repo, local_dir_use_symlinks=False)
     nemo_file = next((os.path.join(folder, f) for f in os.listdir(folder) if f.endswith(".nemo")), None)
     
+    if not nemo_file: raise FileNotFoundError("Fichier .nemo introuvable.")
+
+    # Optimisation RobotsMali : Connecteur flexible pour éviter l'erreur state_dict
     from nemo.core.connectors.save_restore_connector import SaveRestoreConnector
     model = nemo_asr.models.ASRModel.restore_from(
         nemo_file, 
         map_location=torch.device(DEVICE),
         save_restore_connector=SaveRestoreConnector()
     )
+    
     model.to(DEVICE).eval()
-    if DEVICE == "cuda": model.half()
+    if DEVICE == "cuda":
+        model.half()
+        
     _cache[name] = model
     return model
 
+# 3. UTILITAIRES
 def format_srt_time(sec):
     td = time.gmtime(sec)
     ms = int((sec - int(sec)) * 1000)
     return f"{time.strftime('%H:%M:%S', td)},{ms:03}"
 
+# 4. PIPELINE DE TRANSCRIPTION
 def pipeline(video_in, model_name):
     tmp_dir = tempfile.mkdtemp()
     try:
-        if not video_in: return "❌ Source vide", None
+        if not video_in: return "❌ Veuillez sélectionner une vidéo.", None
         
-        yield "⏳ Phase 1 : Extraction Audio...", None
+        # A. Extraction Audio
+        yield "⏳ Phase 1/4 : Extraction audio...", None
         full_wav = os.path.join(tmp_dir, "full.wav")
         subprocess.run(f"ffmpeg -y -i {shlex.quote(video_in)} -vn -ac 1 -ar 16000 {full_wav}", shell=True, check=True)
         
-        yield "⏳ Phase 2 : Segmentation...", None
+        # B. Segmentation
+        yield "⏳ Phase 2/4 : Segmentation (20s)...", None
         subprocess.run(f"ffmpeg -i {full_wav} -f segment -segment_time 20 -c copy {os.path.join(tmp_dir, 'seg_%03d.wav')}", shell=True, check=True)
         audio_segments = sorted(glob.glob(os.path.join(tmp_dir, "seg_*.wav")))
         
-        yield f"⏳ Phase 3 : IA ({model_name})...", None
+        # C. Transcription IA
+        yield f"⏳ Phase 3/4 : Chargement de {model_name}...", None
         model = get_model(model_name)
         
         all_words_ts = []
         for idx, seg_path in enumerate(audio_segments):
             base_time = idx * 20
-            yield f"🎙️ Transcription {idx+1}/{len(audio_segments)}...", None
+            yield f"🎙️ Transcription segment {idx+1}/{len(audio_segments)}...", None
             hyp = model.transcribe([seg_path], return_hypotheses=True)[0]
             if isinstance(hyp, list): hyp = hyp[0]
             text = hyp.text if hasattr(hyp, 'text') else str(hyp)
             words = text.split()
+            
+            # Répartition temporelle
             gap = 20.0 / max(len(words), 1)
             for i, w in enumerate(words):
                 all_words_ts.append({"word": w, "start": base_time + (i * gap), "end": base_time + ((i+1) * gap)})
 
-        yield "⏳ Phase 4 : Rendu Vidéo...", None
+        # D. Génération SRT et Rendu Vidéo
+        yield "⏳ Phase 4/4 : Incrustation sous-titres...", None
         srt_path = os.path.join(tmp_dir, "final.srt")
         with open(srt_path, "w", encoding="utf-8") as f:
             for i in range(0, len(all_words_ts), 6):
@@ -89,38 +122,42 @@ def pipeline(video_in, model_name):
                 f.write(f"{(i//6)+1}\n{format_srt_time(chunk[0]['start'])} --> {format_srt_time(chunk[-1]['end'])}\n")
                 f.write(" ".join([c['word'] for c in chunk]) + "\n\n")
 
-        out_path = os.path.abspath(f"output_{int(time.time())}.mp4")
+        out_path = os.path.abspath(f"robotsmali_result_{int(time.time())}.mp4")
         safe_srt = srt_path.replace("\\", "/").replace(":", "\\:")
-        subprocess.run(f"ffmpeg -y -i {shlex.quote(video_in)} -vf \"subtitles='{safe_srt}':force_style='Alignment=2,FontSize=18'\" -c:v libx264 -preset ultrafast -c:a aac {out_path}", shell=True, check=True)
+        
+        # Style : Jaune, Taille 18, Centré en bas
+        cmd = f"ffmpeg -y -i {shlex.quote(video_in)} -vf \"subtitles='{safe_srt}':force_style='Alignment=2,FontSize=18,PrimaryColour=&H00FFFF'\" -c:v libx264 -preset ultrafast -c:a aac {out_path}"
+        subprocess.run(cmd, shell=True, check=True)
         
         yield "✅ Terminé !", out_path
+
     except Exception as e:
         yield f"❌ Erreur : {str(e)}", None
     finally:
         if os.path.exists(tmp_dir): shutil.rmtree(tmp_dir)
 
-# INTERFACE
+# 5. INTERFACE GRADIO
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.HTML("<h1 style='text-align:center;'>🤖 RobotsMali Transcription</h1>")
+    gr.HTML("<div style='text-align:center;'><h1>🤖 RobotsMali Speech Laboratory</h1><p>Testez nos modèles de transcription et traduction</p></div>")
     
     with gr.Row():
         with gr.Column():
-            v_in = gr.Video(label="Vidéo Source")
-            m_sel = gr.Dropdown(choices=list(MODELS.keys()), value="Soloba V3 (CTC)", label="Modèle")
-            btn = gr.Button("GÉNÉRER", variant="primary")
+            v_input = gr.Video(label="Vidéo")
+            m_input = gr.Dropdown(choices=list(MODELS.keys()), value="Soloba V3 (CTC)", label="Modèle")
+            run_btn = gr.Button("🚀 GÉNÉRER", variant="primary")
             
-            # REVOICI L'EXEMPLE ICI
-            if EXAMPLE_VIDEO:
+            # --- AFFICHAGE DE L'EXEMPLE SI TROUVÉ ---
+            if EXAMPLE_PATH:
                 gr.Examples(
-                    examples=[[EXAMPLE_VIDEO, "Soloba V3 (CTC)"]],
-                    inputs=[v_in, m_sel],
-                    label="Exemple disponible"
+                    examples=[[EXAMPLE_PATH, "Soloba V3 (CTC)"]],
+                    inputs=[v_input, m_input],
+                    label="Vidéo d'exemple"
                 )
 
         with gr.Column():
-            status = gr.Markdown("Prêt.")
-            v_out = gr.Video(label="Résultat")
+            status = gr.Markdown("### État\nPrêt.")
+            v_output = gr.Video(label="Résultat")
 
-    btn.click(pipeline, [v_in, m_sel], [status, v_out])
+    run_btn.click(pipeline, [v_input, m_input], [status, v_output])
 
 demo.launch()