Update app.py

app.py

@@ -2,6 +2,8 @@ import streamlit as st
 import torch
 import tempfile
 import os
+import librosa
+import numpy as np
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
 from peft import PeftModel
 
@@ -40,19 +42,22 @@ if uploaded_file is not None:
 
     # Charger et traiter l'audio
     st.write("📄 **Transcription en cours...**")
-    
-    # Charger l'audio
-    audio_input = processor(temp_audio_path, return_tensors="pt", sampling_rate=16000)
-    input_features = audio_input.input_features
 
-    # Générer la transcription
+    # 🔹 Charger l'audio et convertir en waveform
+    audio, sr = librosa.load(temp_audio_path, sr=16000)  # Whisper attend du 16kHz
+    audio = np.expand_dims(audio, axis=0)  # Ajouter une dimension batch
+
+    # 🔹 Préparer les entrées pour Whisper
+    inputs = processor(audio, sampling_rate=16000, return_tensors="pt")
+
+    # 🔹 Générer la transcription
     with torch.no_grad():
-        predicted_ids = model.generate(input_features)
+        predicted_ids = model.generate(inputs.input_features)
 
-    # Décoder la sortie
+    # 🔹 Décoder la sortie
     transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
-    
-    # Afficher la transcription
+
+    # 🔹 Afficher la transcription
     st.subheader("📝 Transcription :")
     st.text_area("", transcription, height=200)
 
@@ -60,3 +65,4 @@ if uploaded_file is not None:
     os.remove(temp_audio_path)
 
 st.write("🔹 Modèle utilisé :", "Whisper Large + Adaptateur LoRA (SimpleFrog/whisper_finetuned)")
+