Update app.py

app.py

@@ -31,7 +31,7 @@ os.makedirs(TEMP_DIR, exist_ok=True)
 
 def init_metadata_state():
     return []
-
+    
 # Fonction de correction typographique complète avec détection préalable
 def correct_typography(text):
     corrected_text = text
@@ -41,56 +41,58 @@ def correct_typography(text):
     corrected_text = re.sub(r"([?!:;])(?=\w)", r"\1 ", corrected_text)  # Ajout d'un espace après !, ?, : et ; si nécessaire
     corrected_text = re.sub(r"(?<!\d) (\.)", r"\1", corrected_text)  # Suppression de l'espace avant un point
     corrected_text = re.sub(r"(\.) (?=\w)", r". \2", corrected_text)  # Ajout d'un espace après un point si nécessaire
+    
+    # Appliquer la correction uniquement si le texte a changé
     return corrected_text.strip() if corrected_text != text else text
-
 # -------------------------------------------------
-# 2. Transcription de l'audio avec Whisper
+# 2. Transcription de l'audio avec Whisper (Timestamps de fin + Marge de Sécurité)
 # -------------------------------------------------
+
 @spaces.GPU(duration=120)
 def transcribe_audio(audio_path):
-    import os
-    if not audio_path:
-        return "Aucun fichier audio fourni", None, [], "", ""
-    
-    file_name = os.path.basename(audio_path).rsplit('.', 1)[0]  # Extraire le nom sans extension
-    result = pipe(audio_path, return_timestamps="word")
-    words = result.get("chunks", [])
-    
-    if not words:
-        return "Erreur : Aucun timestamp détecté.", None, [], "", file_name
-    
-    raw_transcription = " ".join([w["text"] for w in words])
-    word_timestamps = [(w["text"], w["timestamp"][0], w["timestamp"][1]) for w in words]
-    transcription_with_timestamps = " ".join([f"{w[0]}[{w[1]:.2f}-{w[2]:.2f}]" for w in word_timestamps])
-    
-    return raw_transcription, word_timestamps, transcription_with_timestamps, audio_path, file_name
     if not audio_path:
+        print("[LOG] Aucun fichier audio fourni.")
         return "Aucun fichier audio fourni", None, [], ""
     
+    print(f"[LOG] Début de la transcription de {audio_path}...")
     result = pipe(audio_path, return_timestamps="word")
     words = result.get("chunks", [])
     
     if not words:
+        print("[LOG ERROR] Erreur : Aucun timestamp détecté.")
         return "Erreur : Aucun timestamp détecté.", None, [], ""
     
     raw_transcription = " ".join([w["text"] for w in words])
-    word_timestamps = [(w["text"], w["timestamp"][0], w["timestamp"][1]) for w in words]
-    transcription_with_timestamps = " ".join([f"{w[0]}[{w[1]:.2f}-{w[2]:.2f}]" for w in word_timestamps])
     
-    return raw_transcription, word_timestamps, transcription_with_timestamps, audio_path
+    MARGIN = 0.06  # 60ms
+    word_timestamps = []
+
+    for i, w in enumerate(words):
+        start_time = w["timestamp"][0]
+        end_time = w["timestamp"][1] if w["timestamp"][1] is not None else start_time + 0.5
+
+        if i < len(words) - 1:
+            next_start_time = words[i + 1]["timestamp"][0]
+            end_time = min(end_time + MARGIN, next_start_time - 0.01)
 
+        word_timestamps.append((w["text"], start_time, end_time))
+
+    transcription_with_timestamps = " ".join([f"{w[0]}[{w[1]:.2f}-{w[2]:.2f}]" for w in word_timestamps])
+
+    print(f"[LOG] Transcription brute corrigée : {raw_transcription}")
+    return raw_transcription, word_timestamps, transcription_with_timestamps, audio_path
 
 # -------------------------------------------------
-# 3. Enregistrement des segments définis par l'utilisateur
+# 3. Enregistrement des segments définis par l'utilisateur (Affichage sur Interface)
 # -------------------------------------------------
-def save_segments(table_data, zip_name):
+def save_segments(table_data):
     print("[LOG] Enregistrement des segments définis par l'utilisateur...")
     formatted_data = []
     confirmation_message = "**📌 Segments enregistrés :**\n"
 
     for i, row in table_data.iterrows():
         text, start_time, end_time = row["Texte"], row["Début (s)"], row["Fin (s)"]
-        segment_id = f"{zip_name}_seg_{i+1:02d}"
+        segment_id = f"seg_{i+1:02d}"
 
         try:
             start_time = str(start_time).replace(",", ".")
@@ -114,39 +116,17 @@ def save_segments(table_data, zip_name):
             print(f"[LOG ERROR] Erreur de conversion des timestamps : {e}")
             return pd.DataFrame(), "❌ **Erreur** : Vérifiez que les valeurs sont bien des nombres valides."
 
-    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)", "ID"]), confirmation_message, zip_name
-    formatted_data = []
-    for i, row in table_data.iterrows():
-        formatted_data.append([row["Texte"], float(row["Début (s)"]), float(row["Fin (s)"])] )
-    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)"]), zip_name
+    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)", "ID"]), confirmation_message
 
 # -------------------------------------------------
-# 4. Génération du fichier ZIP avec correction typographique
+# 4. Génération du fichier ZIP
 # -------------------------------------------------
 def generate_zip(metadata_state, audio_path, zip_name):
-    if isinstance(metadata_state, tuple): 
-        metadata_state = metadata_state[0]
-    
     if metadata_state is None or metadata_state.empty:
-        print("[LOG ERROR] Aucun segment valide trouvé pour la génération du ZIP.")
         return None
     
-    zip_folder_name = f"{zip_name}_dataset"
-    zip_path = os.path.join(TEMP_DIR, f"{zip_folder_name}.zip")
-    metadata_csv_path = os.path.join(TEMP_DIR, "metadata.csv")
-    
-    metadata_state["ID"] = [f"{zip_name}_seg_{i+1:02d}" for i in range(len(metadata_state))]
-    metadata_state["Texte"] = metadata_state["Texte"].apply(correct_typography)
-    metadata_state["Commentaires"] = ""
-    metadata_state.to_csv(metadata_csv_path, sep="|", index=False)
-    
-    with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
-        zf.write(metadata_csv_path, "metadata.csv")
-    
-    print("[LOG] Fichier ZIP généré avec succès.")
-    return zip_path
-    if metadata_state is None or metadata_state.empty:
-        return None
+    if not zip_name.strip():
+        zip_name = "processed_audio"
     
     zip_folder_name = f"{zip_name}_dataset"
     zip_path = os.path.join(TEMP_DIR, f"{zip_folder_name}.zip")
@@ -159,6 +139,15 @@ def generate_zip(metadata_state, audio_path, zip_name):
     
     with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
         zf.write(metadata_csv_path, "metadata.csv")
+        original_audio = AudioSegment.from_file(audio_path)
+        
+        for i, row in metadata_state.iterrows():
+            start_ms, end_ms = int(row["Début (s)"] * 1000), int(row["Fin (s)"] * 1000)
+            segment_audio = original_audio[start_ms:end_ms]
+            segment_filename = f"{zip_name}_seg_{i+1:02d}.wav"
+            segment_path = os.path.join(TEMP_DIR, segment_filename)
+            segment_audio.export(segment_path, format="wav")
+            zf.write(segment_path, segment_filename)
     
     return zip_path
 
@@ -168,8 +157,8 @@ def generate_zip(metadata_state, audio_path, zip_name):
 with gr.Blocks() as demo:
     gr.Markdown("# Application de Découpe Audio")
     metadata_state = gr.State(init_metadata_state())
-    zip_name = gr.audio_input = gr.Audio(type="filepath", label="Fichier audio")
-    zip_name = gr.Textbox(label="Nom du fichier ZIP", interactive=True)
+    audio_input = gr.Audio(type="filepath", label="Fichier audio")
+    zip_name = gr.Textbox(label="Nom du fichier ZIP", placeholder="Nom personnalisé", interactive=True)
     raw_transcription = gr.Textbox(label="Transcription", interactive=True)
     transcription_timestamps = gr.Textbox(label="Transcription avec Timestamps", interactive=True)
     table = gr.Dataframe(headers=["Texte", "Début (s)", "Fin (s)"], datatype=["str", "str", "str"], row_count=(1, "dynamic"))
@@ -177,7 +166,7 @@ with gr.Blocks() as demo:
     generate_button = gr.Button("Générer ZIP")
     zip_file = gr.File(label="Télécharger le ZIP")
     
-    audio_input.change(transcribe_audio, inputs=audio_input, outputs=[raw_transcription, word_timestamps, transcription_timestamps, audio_input, zip_name])
+    audio_input.change(transcribe_audio, inputs=audio_input, outputs=[raw_transcription, transcription_timestamps, audio_input, zip_name])
     save_button.click(save_segments, inputs=table, outputs=[metadata_state, zip_name])
     generate_button.click(generate_zip, inputs=[metadata_state, audio_input, zip_name], outputs=zip_file)