Update app.py

app.py

@@ -1,8 +1,10 @@
 import os, warnings, logging, tempfile
 
+# --- Suppression warnings inutile ---
 warnings.filterwarnings("ignore")
 logging.getLogger("nemo_logger").setLevel(logging.ERROR)
 
+# --- CPU compatibility for HF Spaces ---
 os.environ["NEMO_FORCE_CPU"] = "1"
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
@@ -15,12 +17,19 @@ import soundfile as sf
 from moviepy.editor import VideoFileClip, CompositeVideoClip, TextClip
 from nemo.collections import asr as nemo_asr
 
+# --- UNLOCK IMAGEMAGICK POLICY (Required on HF) ---
+for p in ("/etc/ImageMagick/policy.xml", "/etc/ImageMagick-6/policy.xml"):
+    if os.path.exists(p):
+        os.system(
+            'sed -i "s/rights=\\"none\\"/rights=\\"read|write\\"/g" "{}"'.format(p)
+        )
+
 
 # ---------------- CONFIG ---------------- #
 
 SR = 16000
 MAX_VIDEO_BYTES = 200_000_000  # 200MB limite
-TITLE = "RobotsMali Caption Studio — Sous-titrage Automatique en Bambara"
+TITLE = "RobotsMali Caption Studio — Sous-titrage Bambara Automatique"
 
 ASR_MODELS = {
     "Soloba CTC 0.6B V0": "RobotsMali/soloba-ctc-0.6b-v0",
@@ -54,15 +63,15 @@ def extract_audio(video_path, wav_path):
         raise RuntimeError("⚠️ Vidéo trop lourde (>200MB). Compressez puis réessayez.")
 
     os.system(f"ffmpeg -y -i '{video_path}' -ac 1 -ar {SR} -vn '{wav_path}' >/dev/null 2>&1")
-
     audio, sr = sf.read(wav_path)
+
     if sr == 0 or len(audio) == 0:
         raise RuntimeError("⚠️ Audio introuvable ou illisible.")
 
     return len(audio) / sr
 
 
-# ---------------- TRANSCRIBE (UNIFIÉ + SÛR) ---------------- #
+# ---------------- TRANSCRIBE (UNIFIÉ + SAFE) ---------------- #
 
 def transcribe(model, device, wav_path, model_key):
     audio, sr = sf.read(wav_path)
@@ -79,7 +88,7 @@ def transcribe(model, device, wav_path, model_key):
     x = torch.tensor(audio, dtype=torch.float32).unsqueeze(0).to(device)
     ln = torch.tensor([x.shape[1]]).to(device)
 
-    # ---- Priority 1: Soloni precise timestamps ---- #
+    # ---- Soloni: timestamps réels ---- #
     if "Soloni" in model_key and hasattr(model, "decode_and_align"):
         try:
             with torch.no_grad():
@@ -91,16 +100,13 @@ def transcribe(model, device, wav_path, model_key):
                     encoder_output=proc,
                     encoded_lengths=plen
                 )
-
             hyp = hyps[0][0] if isinstance(hyps[0], list) else hyps[0]
-
             if hasattr(hyp, "words") and hyp.words:
                 return [(w.start_offset_ms/1000, w.end_offset_ms/1000, w.word) for w in hyp.words]
-
         except:
-            pass  # fallback auto
+            pass
 
-    # ---- Priority 2: Universal fallback ---- #
+    # ---- Fallback universel (Soloba + QuartzNet + backup Soloni) ---- #
     out = model.transcribe([wav_path])[0]
 
     if hasattr(out, "text"):
@@ -123,6 +129,7 @@ def transcribe(model, device, wav_path, model_key):
         t += d
         if t >= total_s:
             break
+
     return subs
 
 
@@ -141,6 +148,7 @@ def burn(video_path, subs):
             txt = TextClip(
                 w.upper(),
                 fontsize=int(H/20),
+                font="DejaVu-Sans",            # ✅ Police stable
                 color="white",
                 stroke_color="black",
                 stroke_width=2,
@@ -169,20 +177,19 @@ def pipeline(video, model_name, progress=gr.Progress()):
 
     with tempfile.TemporaryDirectory() as td:
         wav = f"{td}/audio.wav"
-
         progress(0.5, "Extraction audio…")
-        duration = extract_audio(video, wav)
+        extract_audio(video, wav)
 
-        progress(0.75, "Transcription en Bambara…")
+        progress(0.75, "Transcription…")
         subs = transcribe(model, device, wav, model_name)
         if not subs:
             return "⚠️ Aucun mot détecté.", None
 
-    progress(0.95, "Incrustation des sous-titres…")
+    progress(0.95, "Incrustation…")
     out = burn(video, subs)
 
     progress(1.0, "✅ Terminé")
-    return f"✅ Sous-titrage terminé avec **{model_name}**", out
+    return f"✅ Sous-titrage terminé ({model_name})", out
 
 
 # ---------------- UI ---------------- #
@@ -194,7 +201,7 @@ h1 { text-align:center; font-weight:800; color:#005BFF; margin-bottom:6px; }
 """
 
 with gr.Blocks(css=CSS, title=TITLE) as demo:
-    gr.Markdown("<h1>RobotsMali Caption Studio</h1><p>Génération automatique de sous-titres en Bambara</p>")
+    gr.Markdown("<h1>RobotsMali Caption Studio</h1><p>Sous-titrage Automatique en Bambara</p>")
     video = gr.File(label="🎥 Importer une vidéo (max 200MB)", type="filepath")
     model = gr.Dropdown(list(ASR_MODELS.keys()), value="Soloni 114M TDT CTC V1", label="🧠 Modèle ASR")
     run = gr.Button("🚀 Générer les sous-titres")