Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,250 +1,1048 @@
|
|
1 |
import streamlit as st
|
2 |
-
|
3 |
-
import base64
|
4 |
import os
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
import tempfile
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
},
|
111 |
-
{
|
112 |
-
"type": "image_url",
|
113 |
-
"image_url": {"url": f"data:{media_type};base64,{base64_image}"},
|
114 |
-
},
|
115 |
-
],
|
116 |
-
}
|
117 |
-
]
|
118 |
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
)
|
124 |
|
125 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
}
|
135 |
-
]
|
136 |
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
147 |
|
148 |
-
|
|
|
|
|
|
|
|
|
149 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
151 |
-
|
152 |
-
st.title("🩺 **Le Radiologue Virtuel**")
|
153 |
-
st.subheader("Une IA avancée pour l'analyse des images médicales")
|
154 |
|
155 |
-
|
156 |
-
|
157 |
-
|
|
|
158 |
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
- **Traitement d'image avancé** : Explorez l'image téléchargée avec des outils comme l'inversion pour une visualisation plus claire.
|
163 |
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
168 |
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
173 |
|
174 |
-
|
175 |
-
|
176 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
-
# Stocker les renseignements cliniques dans la session
|
185 |
-
st.session_state['clinical_info'] = clinical_info
|
186 |
|
187 |
-
|
188 |
-
|
189 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
190 |
|
191 |
-
|
192 |
-
|
193 |
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(uploaded_file.name)[1]) as tmp_file:
|
194 |
tmp_file.write(uploaded_file.getvalue())
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
#
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
from audio_recorder_streamlit import audio_recorder
|
|
|
3 |
import os
|
4 |
+
import msoffcrypto
|
5 |
+
import docx
|
6 |
+
import pptx
|
7 |
+
import pymupdf4llm
|
8 |
import tempfile
|
9 |
+
from typing import List, Optional, Dict, Any
|
10 |
+
from pydub import AudioSegment
|
11 |
+
from groq import Groq
|
12 |
+
from langchain.chains import LLMChain
|
13 |
+
from langchain_groq import ChatGroq
|
14 |
+
from langchain.prompts import PromptTemplate
|
15 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
16 |
+
from langchain.schema import AIMessage, HumanMessage, SystemMessage
|
17 |
+
from datetime import datetime
|
18 |
+
import smtplib
|
19 |
+
from email.mime.text import MIMEText
|
20 |
+
from email.mime.multipart import MIMEMultipart
|
21 |
+
from email.mime.application import MIMEApplication
|
22 |
+
from reportlab.lib import colors
|
23 |
+
from reportlab.lib.pagesizes import letter
|
24 |
+
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer
|
25 |
+
from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle
|
26 |
+
import re
|
27 |
+
from docx import Document
|
28 |
+
from pytube import YouTube
|
29 |
+
from moviepy import VideoFileClip
|
30 |
+
import yt_dlp
|
31 |
+
from youtube_transcript_api import YouTubeTranscriptApi
|
32 |
+
from urllib.parse import urlparse, parse_qs
|
33 |
+
from ratelimit import limits, sleep_and_retry
|
34 |
+
import time
|
35 |
+
import fasttext
|
36 |
+
import requests
|
37 |
+
from requests.auth import HTTPBasicAuth
|
38 |
+
import pikepdf
|
39 |
+
import io
|
40 |
+
import pypdf
|
41 |
+
from PyPDF2 import PdfReader
|
42 |
|
43 |
+
from pptx import Presentation
|
44 |
+
import trafilatura
|
45 |
+
from bs4 import BeautifulSoup
|
46 |
+
|
47 |
+
class Config:
|
48 |
+
"""Centralisation de la configuration"""
|
49 |
+
GROQ_API_KEY = "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"
|
50 |
+
SENDER_EMAIL = "[email protected]"
|
51 |
+
SENDER_PASSWORD = "fkev txsk ldjg nyqs"
|
52 |
+
FASTTEXT_MODEL_PATH = "lid.176.bin"
|
53 |
+
import urllib.request
|
54 |
+
urllib.request.urlretrieve('https://dl.fbaipublicfiles.com/fasttext/supervised-models/lid.176.bin', 'lid.176.bin')
|
55 |
+
|
56 |
+
|
57 |
+
# Classes PDFGenerator et EmailSender restent inchangées...
|
58 |
+
class PDFGenerator:
|
59 |
+
@staticmethod
|
60 |
+
def create_pdf(content: str, filename: str) -> str:
|
61 |
+
doc = SimpleDocTemplate(filename, pagesize=letter)
|
62 |
+
styles = getSampleStyleSheet()
|
63 |
+
custom_style = ParagraphStyle(
|
64 |
+
'CustomStyle',
|
65 |
+
parent=styles['Normal'],
|
66 |
+
spaceBefore=12,
|
67 |
+
spaceAfter=12,
|
68 |
+
fontSize=12,
|
69 |
+
leading=14,
|
70 |
+
)
|
71 |
+
|
72 |
+
story = []
|
73 |
+
title_style = ParagraphStyle(
|
74 |
+
'CustomTitle',
|
75 |
+
parent=styles['Heading1'],
|
76 |
+
fontSize=16,
|
77 |
+
spaceAfter=30,
|
78 |
+
)
|
79 |
+
story.append(Paragraph("Résumé Audio", title_style))
|
80 |
+
story.append(Paragraph(f"Date: {datetime.now().strftime('%d/%m/%Y %H:%M')}", custom_style))
|
81 |
+
story.append(Spacer(1, 20))
|
82 |
+
|
83 |
+
for line in content.split('\n'):
|
84 |
+
if line.strip():
|
85 |
+
if line.startswith('#'):
|
86 |
+
story.append(Paragraph(line.strip('# '), styles['Heading2']))
|
87 |
+
else:
|
88 |
+
story.append(Paragraph(line, custom_style))
|
89 |
+
|
90 |
+
doc.build(story)
|
91 |
+
return filename
|
92 |
+
|
93 |
+
class EmailSender:
|
94 |
+
def __init__(self, sender_email: str, sender_password: str):
|
95 |
+
self.sender_email = Config.SENDER_EMAIL
|
96 |
+
self.sender_password = Config.SENDER_PASSWORD
|
97 |
+
|
98 |
+
def send_email(self, recipient_email: str, subject: str, body: str, pdf_path: str) -> bool:
|
99 |
+
try:
|
100 |
+
msg = MIMEMultipart()
|
101 |
+
msg['From'] = self.sender_email
|
102 |
+
msg['To'] = recipient_email
|
103 |
+
msg['Subject'] = subject
|
104 |
+
msg.attach(MIMEText(body, 'plain'))
|
105 |
+
|
106 |
+
with open(pdf_path, 'rb') as f:
|
107 |
+
pdf_attachment = MIMEApplication(f.read(), _subtype='pdf')
|
108 |
+
pdf_attachment.add_header('Content-Disposition', 'attachment', filename=os.path.basename(pdf_path))
|
109 |
+
msg.attach(pdf_attachment)
|
110 |
+
|
111 |
+
server = smtplib.SMTP('smtp.gmail.com', 587)
|
112 |
+
server.starttls()
|
113 |
+
server.login(self.sender_email, self.sender_password)
|
114 |
+
server.send_message(msg)
|
115 |
+
server.quit()
|
116 |
+
return True
|
117 |
+
except Exception as e:
|
118 |
+
st.error(f"Erreur d'envoi d'email: {str(e)}")
|
119 |
+
return False
|
120 |
+
|
121 |
+
class AudioProcessor:
|
122 |
+
def __init__(self, model_name: str, prompt: str = None, chunk_length_ms: int = 300000):
|
123 |
+
self.chunk_length_ms = chunk_length_ms
|
124 |
+
self.groq_client = Groq(api_key=Config.GROQ_API_KEY)
|
125 |
+
self.llm = ChatGroq(
|
126 |
+
model=model_name,
|
127 |
+
temperature=0,
|
128 |
+
api_key=Config.GROQ_API_KEY
|
129 |
+
)
|
130 |
+
self.custom_prompt = prompt
|
131 |
+
self.language_detector = fasttext.load_model(Config.FASTTEXT_MODEL_PATH)
|
132 |
+
self.text_splitter = RecursiveCharacterTextSplitter(
|
133 |
+
chunk_size=4000,
|
134 |
+
chunk_overlap=200
|
135 |
+
)
|
136 |
+
#self.custom_prompt = prompt
|
137 |
+
# Définition des limites de taux : 5000 tokens par minute
|
138 |
+
self.CALLS_PER_MINUTE = 5000
|
139 |
+
self.PERIOD = 60 # 60 secondes = 1 minute
|
140 |
+
# Add language detection model
|
141 |
+
#self.language_detector = fasttext.load_model('lid.176.bin')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
|
143 |
+
def check_language(self, text: str) -> str:
|
144 |
+
"""Vérifie si le texte est en français"""
|
145 |
+
prediction = self.language_detector.predict(text.replace('\n', ' '))
|
146 |
+
return "OUI" if prediction[0][0] == '__label__fr' else "NON"
|
|
|
147 |
|
148 |
+
def translate_to_french(self, text: str) -> str:
|
149 |
+
"""Traduit le texte en français si nécessaire"""
|
150 |
+
try:
|
151 |
+
messages = [
|
152 |
+
SystemMessage(content="Vous êtes un traducteur professionnel. Traduisez le texte suivant en français en conservant le format et la structure:"),
|
153 |
+
HumanMessage(content=text)
|
154 |
+
]
|
155 |
+
result = self._make_api_call(messages)
|
156 |
+
return result.generations[0][0].text
|
157 |
+
except Exception as e:
|
158 |
+
if "rate_limit_exceeded" in str(e):
|
159 |
+
time.sleep(60)
|
160 |
+
return self.translate_to_french(text)
|
161 |
+
raise e
|
162 |
|
163 |
+
@sleep_and_retry
|
164 |
+
@limits(calls=5000, period=60)
|
165 |
+
def _make_api_call(self, messages):
|
166 |
+
return self.llm.generate([messages])
|
167 |
+
|
168 |
+
def chunk_audio(self, file_path: str) -> List[AudioSegment]:
|
169 |
+
try:
|
170 |
+
audio = AudioSegment.from_file(file_path)
|
171 |
+
if len(audio) < self.chunk_length_ms:
|
172 |
+
return [audio]
|
173 |
+
return [
|
174 |
+
audio[i:i + self.chunk_length_ms]
|
175 |
+
for i in range(0, len(audio), self.chunk_length_ms)
|
176 |
+
]
|
177 |
+
except Exception as e:
|
178 |
+
st.error(f"Error processing audio file: {str(e)}")
|
179 |
+
return []
|
180 |
+
|
181 |
+
def transcribe_chunk(self, audio_chunk: AudioSegment) -> str:
|
182 |
+
try:
|
183 |
+
with tempfile.NamedTemporaryFile(suffix='.mp3', delete=False) as temp_file:
|
184 |
+
audio_chunk.export(temp_file.name, format="mp3")
|
185 |
+
with open(temp_file.name, "rb") as audio_file:
|
186 |
+
try:
|
187 |
+
response = self.groq_client.audio.transcriptions.create(
|
188 |
+
file=audio_file,
|
189 |
+
model="whisper-large-v3-turbo",
|
190 |
+
language="fr"
|
191 |
+
)
|
192 |
+
except Exception as e:
|
193 |
+
if "rate_limit_exceeded" in str(e):
|
194 |
+
st.warning("Limite de taux atteinte pendant la transcription. Attente avant nouvelle tentative...")
|
195 |
+
time.sleep(60)
|
196 |
+
return self.transcribe_chunk(audio_chunk)
|
197 |
+
raise e
|
198 |
+
os.unlink(temp_file.name)
|
199 |
+
return response.text
|
200 |
+
except Exception as e:
|
201 |
+
st.error(f"Transcription error: {str(e)}")
|
202 |
+
return ""
|
203 |
+
|
204 |
+
|
205 |
+
# Dans la classe AudioProcessor, ajoutez cette méthode :
|
206 |
+
def split_text(self, text: str, max_tokens: int = 4000) -> List[str]:
|
207 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
208 |
+
chunk_size=max_tokens * 4, # Estimation approximative tokens -> caractères
|
209 |
+
chunk_overlap=200,
|
210 |
+
length_function=len,
|
211 |
+
separators=["\n\n", "\n", " ", ""]
|
212 |
+
)
|
213 |
+
return text_splitter.split_text(text)
|
214 |
+
|
215 |
+
def generate_summary(self, transcription: str) -> str:
|
216 |
+
default_prompt = """
|
217 |
+
Vous êtes un assistant expert spécialisé dans le résumé et l'analyse d'enregistrements audio en langue française.
|
218 |
+
Voici la transcription à analyser:
|
219 |
+
|
220 |
+
{transcript}
|
221 |
+
|
222 |
+
Veuillez fournir:
|
223 |
+
1. Un résumé concis (3-4 phrases)
|
224 |
+
2. Les points clés (maximum 5 points)
|
225 |
+
3. Les actions recommandées (si pertinent)
|
226 |
+
4. Une conclusion brève
|
227 |
+
|
228 |
+
Format souhaité:
|
229 |
+
# Résumé
|
230 |
+
[votre résumé]
|
231 |
+
|
232 |
+
# Points Clés
|
233 |
+
• [point 1]
|
234 |
+
• [point 2]
|
235 |
+
...
|
236 |
+
|
237 |
+
# Actions Recommandées
|
238 |
+
1. [action 1]
|
239 |
+
2. [action 2]
|
240 |
+
...
|
241 |
+
|
242 |
+
# Conclusion
|
243 |
+
[votre conclusion]
|
244 |
+
"""
|
245 |
+
|
246 |
+
prompt_template = self.custom_prompt if self.custom_prompt else default_prompt
|
247 |
+
|
248 |
+
try:
|
249 |
+
chain = LLMChain(
|
250 |
+
llm=self.llm,
|
251 |
+
prompt=PromptTemplate(
|
252 |
+
template=prompt_template,
|
253 |
+
input_variables=["transcript"]
|
254 |
+
)
|
255 |
+
)
|
256 |
+
|
257 |
+
summary = chain.run(transcript=transcription)
|
258 |
+
|
259 |
+
# Vérification de la langue
|
260 |
+
if self.check_language(summary) == "NON":
|
261 |
+
st.warning("Résumé généré dans une autre langue. Traduction en cours...")
|
262 |
+
summary = self.translate_to_french(summary)
|
263 |
+
|
264 |
+
return summary
|
265 |
+
except Exception as e:
|
266 |
+
if "rate_limit_exceeded" in str(e):
|
267 |
+
st.warning("Limite de taux atteinte. Attente avant nouvelle tentative...")
|
268 |
+
time.sleep(60) # Attendre 1 minute
|
269 |
+
return self.generate_summary(transcription)
|
270 |
+
raise e
|
271 |
|
272 |
+
# Méthodes existantes inchangées...
|
273 |
+
|
274 |
+
|
275 |
+
|
276 |
+
def summarize_long_transcription(self, transcription: str) -> str:
|
277 |
+
chunks = self.split_text(transcription, max_tokens=4000)
|
278 |
+
partial_summaries = []
|
279 |
+
|
280 |
+
for i, chunk in enumerate(chunks):
|
281 |
+
st.write(f"Traitement du segment {i + 1}/{len(chunks)}...")
|
282 |
+
try:
|
283 |
+
messages = [
|
284 |
+
SystemMessage(content="Vous êtes un assistant expert en résumé de texte en français."),
|
285 |
+
HumanMessage(content=f"Résumez ce texte en français : {chunk}")
|
286 |
+
]
|
287 |
+
result = self._make_api_call(messages)
|
288 |
+
partial_summary = result.generations[0][0].text
|
289 |
+
|
290 |
+
# Vérification de la langue pour chaque segment
|
291 |
+
if self.check_language(partial_summary) == "NON":
|
292 |
+
partial_summary = self.translate_to_french(partial_summary)
|
293 |
+
|
294 |
+
partial_summaries.append(partial_summary)
|
295 |
+
except Exception as e:
|
296 |
+
if "rate_limit_exceeded" in str(e):
|
297 |
+
st.warning(f"Limite de taux atteinte au segment {i+1}. Attente avant nouvelle tentative...")
|
298 |
+
time.sleep(60)
|
299 |
+
i -= 1
|
300 |
+
continue
|
301 |
+
raise e
|
302 |
+
|
303 |
+
try:
|
304 |
+
final_prompt = f"""Combinez ces résumés partiels en un résumé global cohérent en langue française :
|
305 |
+
|
306 |
+
{' '.join(partial_summaries)}
|
307 |
+
"""
|
308 |
+
messages = [
|
309 |
+
SystemMessage(content="Vous êtes un assistant expert en résumé de texte en français."),
|
310 |
+
HumanMessage(content=final_prompt)
|
311 |
+
]
|
312 |
+
final_result = self._make_api_call(messages)
|
313 |
+
final_summary = final_result.generations[0][0].text
|
314 |
+
|
315 |
+
# Vérification finale de la langue
|
316 |
+
if self.check_language(final_summary) == "NON":
|
317 |
+
st.warning("Résumé final dans une autre langue. Traduction en cours...")
|
318 |
+
final_summary = self.translate_to_french(final_summary)
|
319 |
+
|
320 |
+
return final_summary
|
321 |
+
|
322 |
+
except Exception as e:
|
323 |
+
if "rate_limit_exceeded" in str(e):
|
324 |
+
st.warning("Limite de taux atteinte lors de la génération du résumé final. Attente avant nouvelle tentative...")
|
325 |
+
time.sleep(60)
|
326 |
+
return self.summarize_long_transcription(transcription)
|
327 |
+
raise e
|
328 |
+
"""def summarize_long_transcription(self, transcription: str) -> str:
|
329 |
+
try:
|
330 |
+
chunks = self.split_text(transcription)
|
331 |
+
partial_summaries = []
|
332 |
+
|
333 |
+
for i, chunk in enumerate(chunks):
|
334 |
+
st.write(f"Traitement du segment {i + 1}/{len(chunks)}...")
|
335 |
+
summary = self._process_chunk(chunk)
|
336 |
+
partial_summaries.append(summary)
|
337 |
+
|
338 |
+
return self._combine_summaries(partial_summaries)
|
339 |
+
except Exception as e:
|
340 |
+
if "rate_limit_exceeded" in str(e):
|
341 |
+
time.sleep(60)
|
342 |
+
return self.summarize_long_transcription(transcription)
|
343 |
+
raise e
|
344 |
+
|
345 |
+
def _process_chunk(self, chunk: str) -> str:
|
346 |
+
messages = [
|
347 |
+
SystemMessage(content="Résumez ce texte en français :"),
|
348 |
+
HumanMessage(content=chunk)
|
349 |
+
]
|
350 |
+
result = self._make_api_call(messages)
|
351 |
+
summary = result.generations[0][0].text
|
352 |
+
|
353 |
+
if self.check_language(summary) == "NON":
|
354 |
+
summary = self.translate_to_french(summary)
|
355 |
+
|
356 |
+
return summary
|
357 |
+
|
358 |
+
def _combine_summaries(self, summaries: List[str]) -> str:
|
359 |
+
try:
|
360 |
+
messages = [
|
361 |
+
SystemMessage(content="Combinez ces résumés en un résumé global cohérent en français :"),
|
362 |
+
HumanMessage(content=' '.join(summaries))
|
363 |
+
]
|
364 |
+
result = self._make_api_call(messages)
|
365 |
+
final_summary = result.generations[0][0].text
|
366 |
+
|
367 |
+
if self.check_language(final_summary) == "NON":
|
368 |
+
final_summary = self.translate_to_french(final_summary)
|
369 |
+
|
370 |
+
return final_summary
|
371 |
+
except Exception as e:
|
372 |
+
if "rate_limit_exceeded" in str(e):
|
373 |
+
time.sleep(60)
|
374 |
+
return self._combine_summaries(summaries)
|
375 |
+
raise e"""
|
376 |
+
|
377 |
+
class VideoProcessor:
|
378 |
+
def __init__(self):
|
379 |
+
self.supported_formats = ['.mp4', '.avi', '.mov', '.mkv']
|
380 |
+
self.ydl_opts = {
|
381 |
+
'format': 'bestaudio/best',
|
382 |
+
'postprocessors': [{
|
383 |
+
'key': 'FFmpegExtractAudio',
|
384 |
+
'preferredcodec': 'mp3',
|
385 |
+
'preferredquality': '192',
|
386 |
+
}],
|
387 |
+
'outtmpl': 'temp_audio.%(ext)s'
|
388 |
}
|
|
|
389 |
|
390 |
+
def extract_video_id(self, url: str) -> str:
|
391 |
+
try:
|
392 |
+
parsed_url = urlparse(url)
|
393 |
+
if parsed_url.hostname in ['www.youtube.com', 'youtube.com']:
|
394 |
+
return parse_qs(parsed_url.query)['v'][0]
|
395 |
+
elif parsed_url.hostname == 'youtu.be':
|
396 |
+
return parsed_url.path[1:]
|
397 |
+
return None
|
398 |
+
except Exception:
|
399 |
+
return None
|
400 |
+
|
401 |
+
def get_youtube_transcription(self, video_id: str) -> Optional[str]:
|
402 |
+
try:
|
403 |
+
transcript_list = YouTubeTranscriptApi.get_transcript(video_id, languages=['fr', 'en'])
|
404 |
+
return ' '.join(entry['text'] for entry in transcript_list)
|
405 |
+
except Exception:
|
406 |
+
return None
|
407 |
+
|
408 |
+
def download_youtube_audio(self, url: str) -> str:
|
409 |
+
with yt_dlp.YoutubeDL(self.ydl_opts) as ydl:
|
410 |
+
ydl.download([url])
|
411 |
+
return 'temp_audio.mp3'
|
412 |
+
|
413 |
+
def extract_audio_from_video(self, video_path: str) -> str:
|
414 |
+
try:
|
415 |
+
audio_path = f"{os.path.splitext(video_path)[0]}.mp3"
|
416 |
+
with VideoFileClip(video_path) as video:
|
417 |
+
video.audio.write_audiofile(audio_path)
|
418 |
+
return audio_path
|
419 |
+
except Exception as e:
|
420 |
+
st.error(f"Erreur lors de l'extraction audio: {str(e)}")
|
421 |
+
raise
|
422 |
+
|
423 |
+
class DocumentProcessor:
|
424 |
+
def __init__(self, model_name: str, prompt: str = None):
|
425 |
+
self.llm = ChatGroq(
|
426 |
+
model=model_name,
|
427 |
+
temperature=0,
|
428 |
+
api_key=Config.GROQ_API_KEY
|
429 |
+
)
|
430 |
+
self.custom_prompt = prompt
|
431 |
+
#self.text_splitter = RecursiveCharacterTextSplitter(
|
432 |
+
# chunk_size=4000,
|
433 |
+
# chunk_overlap=200
|
434 |
+
#)
|
435 |
+
self.language_detector = fasttext.load_model('lid.176.bin')
|
436 |
+
|
437 |
+
def split_text(self, text: str, max_tokens: int = 4000) -> List[str]:
|
438 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
439 |
+
chunk_size=max_tokens * 4, # Estimation approximative tokens -> caractères
|
440 |
+
chunk_overlap=200,
|
441 |
+
length_function=len,
|
442 |
+
separators=["\n\n", "\n", " ", ""]
|
443 |
+
)
|
444 |
+
return text_splitter.split_text(text)
|
445 |
+
|
446 |
+
def check_language(self, text: str) -> str:
|
447 |
+
"""Vérifie si le texte est en français"""
|
448 |
+
prediction = self.language_detector.predict(text.replace('\n', ' '))
|
449 |
+
return "OUI" if prediction[0][0] == '__label__fr' else "NON"
|
450 |
|
451 |
+
def translate_to_french(self, text: str) -> str:
|
452 |
+
"""Traduit le texte en français si nécessaire"""
|
453 |
+
try:
|
454 |
+
messages = [
|
455 |
+
SystemMessage(content="Vous êtes un traducteur professionnel. Traduisez le texte suivant en français en conservant le format et la structure:"),
|
456 |
+
HumanMessage(content=text)
|
457 |
+
]
|
458 |
+
result = self._make_api_call(messages)
|
459 |
+
return result.generations[0][0].text
|
460 |
+
except Exception as e:
|
461 |
+
if "rate_limit_exceeded" in str(e):
|
462 |
+
time.sleep(60)
|
463 |
+
return self.translate_to_french(text)
|
464 |
+
raise e
|
465 |
|
466 |
+
# Méthodes existantes de DocumentProcessor inchangées...
|
467 |
+
@sleep_and_retry
|
468 |
+
@limits(calls=5000, period=60)
|
469 |
+
def _make_api_call(self, messages):
|
470 |
+
return self.llm.generate([messages])
|
471 |
|
472 |
+
|
473 |
+
|
474 |
+
def process_protected_pdf(self, file_path: str, password: str = None) -> str:
|
475 |
+
"""
|
476 |
+
Traite un PDF, avec ou sans mot de passe, et extrait le texte.
|
477 |
+
|
478 |
+
:param file_path: Chemin vers le fichier PDF.
|
479 |
+
:param password: Mot de passe du fichier PDF (si nécessaire).
|
480 |
+
:return: Texte extrait du PDF.
|
481 |
+
"""
|
482 |
+
try:
|
483 |
+
# Si un mot de passe est fourni, tenter de déverrouiller le PDF
|
484 |
+
if password:
|
485 |
+
with pikepdf.open(file_path, password=password) as pdf:
|
486 |
+
unlocked_pdf_path = "unlocked_temp.pdf"
|
487 |
+
pdf.save(unlocked_pdf_path)
|
488 |
+
|
489 |
+
# Utiliser le fichier temporaire déverrouillé
|
490 |
+
reader = PdfReader(unlocked_pdf_path)
|
491 |
+
text = ""
|
492 |
+
for page in reader.pages:
|
493 |
+
text += page.extract_text()
|
494 |
+
|
495 |
+
# Supprimer le fichier temporaire
|
496 |
+
os.remove(unlocked_pdf_path)
|
497 |
+
|
498 |
+
else:
|
499 |
+
# Si aucun mot de passe, traiter directement le PDF
|
500 |
+
reader = PdfReader(file_path)
|
501 |
+
text = ""
|
502 |
+
for page in reader.pages:
|
503 |
+
text += page.extract_text()
|
504 |
|
505 |
+
return text
|
|
|
|
|
506 |
|
507 |
+
except pikepdf.PasswordError:
|
508 |
+
raise ValueError("Mot de passe PDF incorrect")
|
509 |
+
except Exception as e:
|
510 |
+
raise RuntimeError(f"Erreur lors du traitement du PDF : {e}")
|
511 |
|
512 |
+
def process_protected_office(self, file, file_type: str, password: str = None) -> str:
|
513 |
+
"""
|
514 |
+
Traite un fichier Office (protégé ou non) et extrait le texte.
|
|
|
515 |
|
516 |
+
:param file: Le fichier Office à traiter.
|
517 |
+
:param password: Mot de passe du fichier (si nécessaire, sinon None).
|
518 |
+
:param file_type: Type du fichier ('docx' ou 'pptx').
|
519 |
+
:return: Texte extrait du fichier.
|
520 |
+
"""
|
521 |
+
try:
|
522 |
+
if password:
|
523 |
+
# Cas où un mot de passe est fourni, tenter de déverrouiller le fichier
|
524 |
+
office_file = msoffcrypto.OfficeFile(file)
|
525 |
+
office_file.load_key(password=password)
|
526 |
+
|
527 |
+
decrypted = io.BytesIO()
|
528 |
+
office_file.decrypt(decrypted)
|
529 |
+
|
530 |
+
if file_type == 'docx':
|
531 |
+
doc = docx.Document(decrypted)
|
532 |
+
return "\n".join([p.text for p in doc.paragraphs])
|
533 |
+
elif file_type == 'pptx':
|
534 |
+
ppt = pptx.Presentation(decrypted)
|
535 |
+
return "\n".join([shape.text for slide in ppt.slides
|
536 |
+
for shape in slide.shapes if hasattr(shape, "text")])
|
537 |
+
else:
|
538 |
+
# Cas où aucun mot de passe n'est fourni, traiter directement le fichier
|
539 |
+
if file_type == 'docx':
|
540 |
+
doc = docx.Document(file) # Charger le fichier sans décryptage
|
541 |
+
return "\n".join([p.text for p in doc.paragraphs])
|
542 |
+
elif file_type == 'pptx':
|
543 |
+
ppt = pptx.Presentation(file)
|
544 |
+
return "\n".join([shape.text for slide in ppt.slides
|
545 |
+
for shape in slide.shapes if hasattr(shape, "text")])
|
546 |
+
|
547 |
+
raise ValueError("Type de fichier non supporté. Utilisez 'docx' ou 'pptx'.")
|
548 |
+
|
549 |
+
except msoffcrypto.exceptions.InvalidKeyError:
|
550 |
+
raise ValueError("Mot de passe incorrect ou fichier non valide.")
|
551 |
+
except Exception as e:
|
552 |
+
raise RuntimeError(f"Erreur lors du traitement du fichier Office : {e}")
|
553 |
|
554 |
+
"""def process_protected_office(self, file, password: str=None, file_type: str) -> str:
|
555 |
+
try:
|
556 |
+
office_file = msoffcrypto.OfficeFile(file)
|
557 |
+
office_file.load_key(password=password)
|
558 |
+
|
559 |
+
decrypted = io.BytesIO()
|
560 |
+
office_file.decrypt(decrypted)
|
561 |
+
|
562 |
+
if file_type == 'docx':
|
563 |
+
doc = docx.Document(decrypted)
|
564 |
+
return "\n".join([p.text for p in doc.paragraphs])
|
565 |
+
elif file_type == 'pptx':
|
566 |
+
ppt = pptx.Presentation(decrypted)
|
567 |
+
return "\n".join([shape.text for slide in ppt.slides
|
568 |
+
for shape in slide.shapes if hasattr(shape, "text")])
|
569 |
+
except Exception:
|
570 |
+
raise ValueError("Mot de passe document incorrect")"""
|
571 |
|
572 |
+
def scrape_web_content(self, url: str, auth: Dict[str, str] = None) -> str:
|
573 |
+
try:
|
574 |
+
if auth:
|
575 |
+
session = requests.Session()
|
576 |
+
session.auth = HTTPBasicAuth(auth['username'], auth['password'])
|
577 |
+
response = session.get(url, timeout=30)
|
578 |
+
else:
|
579 |
+
response = requests.get(url, timeout=30)
|
580 |
+
|
581 |
+
response.raise_for_status()
|
582 |
+
downloaded = trafilatura.extract(response.text)
|
583 |
+
|
584 |
+
if not downloaded:
|
585 |
+
raise ValueError("Impossible d'extraire le contenu de cette page")
|
586 |
+
return downloaded
|
587 |
+
|
588 |
+
except requests.exceptions.HTTPError as e:
|
589 |
+
if e.response.status_code == 401:
|
590 |
+
raise ValueError("Authentification requise pour accéder à cette page")
|
591 |
+
elif e.response.status_code == 404:
|
592 |
+
raise ValueError("Page introuvable")
|
593 |
+
else:
|
594 |
+
raise ValueError(f"Erreur HTTP: {e.response.status_code}")
|
595 |
+
except requests.exceptions.RequestException:
|
596 |
+
raise ValueError("URL invalide ou inaccessible")
|
597 |
|
598 |
+
def summarize_text(self, transcription: str) -> str:
|
599 |
+
chunks = self.split_text(transcription, max_tokens=4000)
|
600 |
+
partial_summaries = []
|
601 |
+
|
602 |
+
for i, chunk in enumerate(chunks):
|
603 |
+
st.write(f"Traitement du segment {i + 1}/{len(chunks)}...")
|
604 |
+
try:
|
605 |
+
messages = [
|
606 |
+
SystemMessage(content="Vous êtes un assistant expert en résumé de texte en français."),
|
607 |
+
HumanMessage(content=f"Résumez ce texte en français : {chunk}")
|
608 |
+
]
|
609 |
+
result = self._make_api_call(messages)
|
610 |
+
partial_summary = result.generations[0][0].text
|
611 |
+
|
612 |
+
# Vérification de la langue pour chaque segment
|
613 |
+
if self.check_language(partial_summary) == "NON":
|
614 |
+
partial_summary = self.translate_to_french(partial_summary)
|
615 |
+
|
616 |
+
partial_summaries.append(partial_summary)
|
617 |
+
except Exception as e:
|
618 |
+
if "rate_limit_exceeded" in str(e):
|
619 |
+
st.warning(f"Limite de taux atteinte au segment {i+1}. Attente avant nouvelle tentative...")
|
620 |
+
time.sleep(60)
|
621 |
+
i -= 1
|
622 |
+
continue
|
623 |
+
raise e
|
624 |
+
|
625 |
+
try:
|
626 |
+
final_prompt = f"""Combinez ces résumés partiels en un résumé global cohérent en langue française :
|
627 |
+
|
628 |
+
{' '.join(partial_summaries)}
|
629 |
+
"""
|
630 |
+
messages = [
|
631 |
+
SystemMessage(content="Vous êtes un assistant expert en résumé de texte en français."),
|
632 |
+
HumanMessage(content=final_prompt)
|
633 |
+
]
|
634 |
+
final_result = self._make_api_call(messages)
|
635 |
+
final_summary = final_result.generations[0][0].text
|
636 |
+
|
637 |
+
# Vérification finale de la langue
|
638 |
+
if self.check_language(final_summary) == "NON":
|
639 |
+
st.warning("Résumé final dans une autre langue. Traduction en cours...")
|
640 |
+
final_summary = self.translate_to_french(final_summary)
|
641 |
+
|
642 |
+
return final_summary
|
643 |
+
|
644 |
+
except Exception as e:
|
645 |
+
if "rate_limit_exceeded" in str(e):
|
646 |
+
st.warning("Limite de taux atteinte lors de la génération du résumé final. Attente avant nouvelle tentative...")
|
647 |
+
time.sleep(60)
|
648 |
+
return self.summarize_long_transcription(transcription)
|
649 |
+
raise e
|
650 |
+
"""def summarize_text(self, text: str) -> str:
|
651 |
+
try:
|
652 |
+
chunks = self.text_splitter.split_text(text)
|
653 |
+
summaries = []
|
654 |
+
|
655 |
+
for chunk in chunks:
|
656 |
+
messages = [
|
657 |
+
SystemMessage(content="Générez un résumé en français de ce texte:"),
|
658 |
+
HumanMessage(content=chunk)
|
659 |
+
]
|
660 |
+
summary = self._make_api_call(messages).generations[0][0].text
|
661 |
+
summaries.append(summary)
|
662 |
+
|
663 |
+
if len(summaries) > 1:
|
664 |
+
final_messages = [
|
665 |
+
SystemMessage(content="Combinez ces résumés en un résumé cohérent en français:"),
|
666 |
+
HumanMessage(content="\n".join(summaries))
|
667 |
+
]
|
668 |
+
final_summary = self._make_api_call(final_messages).generations[0][0].text
|
669 |
+
return final_summary
|
670 |
+
|
671 |
+
return summaries[0]
|
672 |
+
|
673 |
+
except Exception as e:
|
674 |
+
raise ValueError(f"Erreur de traitement: {str(e)}")"""
|
675 |
+
|
676 |
+
def generate_docx(content: str, filename: str):
|
677 |
+
doc = Document()
|
678 |
+
doc.add_heading('Résumé Audio', 0)
|
679 |
+
doc.add_paragraph(f"Date: {datetime.now().strftime('%d/%m/%Y %H:%M')}")
|
680 |
+
|
681 |
+
for line in content.split('\n'):
|
682 |
+
if line.strip():
|
683 |
+
if line.startswith('#'):
|
684 |
+
doc.add_heading(line.strip('# '), level=1)
|
685 |
+
else:
|
686 |
+
doc.add_paragraph(line)
|
687 |
+
|
688 |
+
doc.save(filename)
|
689 |
+
return filename
|
690 |
|
|
|
|
|
691 |
|
692 |
+
def model_selection_sidebar():
|
693 |
+
"""Configuration du modèle dans la barre latérale"""
|
694 |
+
with st.sidebar:
|
695 |
+
st.title("Configuration")
|
696 |
+
model = st.selectbox(
|
697 |
+
"Sélectionnez un modèle",
|
698 |
+
[
|
699 |
+
"mixtral-8x7b-32768",
|
700 |
+
"llama-3.3-70b-versatile",
|
701 |
+
"gemma2-9b-i",
|
702 |
+
"llama3-70b-8192"
|
703 |
+
]
|
704 |
+
)
|
705 |
+
prompt = st.text_area(
|
706 |
+
"Instructions personnalisées pour le résumé",
|
707 |
+
placeholder="Ex: Résumé de réunion avec points clés et actions"
|
708 |
+
)
|
709 |
+
return model, prompt
|
710 |
|
711 |
+
def save_uploaded_file(uploaded_file) -> str:
|
712 |
+
"""Sauvegarde un fichier uploadé et retourne son chemin"""
|
713 |
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(uploaded_file.name)[1]) as tmp_file:
|
714 |
tmp_file.write(uploaded_file.getvalue())
|
715 |
+
return tmp_file.name
|
716 |
+
|
717 |
+
def is_valid_email(email: str) -> bool:
|
718 |
+
"""Valide le format d'une adresse email"""
|
719 |
+
pattern = r'^[\w\.-]+@[\w\.-]+\.\w+$'
|
720 |
+
return bool(re.match(pattern, email))
|
721 |
+
|
722 |
+
def enhance_main():
|
723 |
+
"""Fonction principale avec gestion des états et des erreurs améliorée"""
|
724 |
+
st.set_page_config(page_title="Multimodal Content Summarizer", page_icon="📝")
|
725 |
+
|
726 |
+
if "audio_processor" not in st.session_state:
|
727 |
+
model_name, custom_prompt = model_selection_sidebar()
|
728 |
+
st.session_state.audio_processor = AudioProcessor(model_name, custom_prompt)
|
729 |
+
|
730 |
+
if "auth_required" not in st.session_state:
|
731 |
+
st.session_state.auth_required = False
|
732 |
+
|
733 |
+
# Interface principale
|
734 |
+
source_type = st.radio("Type de source", ["Audio/Vidéo", "Document", "Web"])
|
735 |
+
|
736 |
+
try:
|
737 |
+
if source_type == "Audio/Vidéo":
|
738 |
+
process_audio_video()
|
739 |
+
elif source_type == "Document":
|
740 |
+
process_document()
|
741 |
+
else: # Web
|
742 |
+
process_web()
|
743 |
+
except Exception as e:
|
744 |
+
st.error(f"Une erreur est survenue: {str(e)}")
|
745 |
+
st.error("Veuillez réessayer ou contacter le support.")
|
746 |
+
|
747 |
+
def process_audio_video():
|
748 |
+
"""Traitement des sources audio et vidéo"""
|
749 |
+
source = st.radio("Choisissez votre source", ["Audio", "Vidéo locale", "YouTube"])
|
750 |
+
|
751 |
+
if source == "Audio":
|
752 |
+
handle_audio_input()
|
753 |
+
elif source == "Vidéo locale":
|
754 |
+
handle_video_input()
|
755 |
+
else: # YouTube
|
756 |
+
handle_youtube_input()
|
757 |
+
|
758 |
+
def handle_audio_input():
|
759 |
+
"""Gestion des entrées audio"""
|
760 |
+
uploaded_file = st.file_uploader("Fichier audio", type=['mp3', 'wav', 'm4a', 'ogg'])
|
761 |
+
audio_bytes = audio_recorder()
|
762 |
+
|
763 |
+
if uploaded_file or audio_bytes:
|
764 |
+
process_and_display_results(uploaded_file, audio_bytes)
|
765 |
+
|
766 |
+
def handle_video_input():
|
767 |
+
"""Gestion des entrées vidéo"""
|
768 |
+
uploaded_video = st.file_uploader("Fichier vidéo", type=['mp4', 'avi', 'mov', 'mkv'])
|
769 |
+
if uploaded_video:
|
770 |
+
st.video(uploaded_video)
|
771 |
+
with st.spinner("Extraction de l'audio..."):
|
772 |
+
video_processor = VideoProcessor()
|
773 |
+
video_path = save_uploaded_file(uploaded_video)
|
774 |
+
audio_path = video_processor.extract_audio_from_video(video_path)
|
775 |
+
process_and_display_results(audio_path)
|
776 |
+
|
777 |
+
def handle_youtube_input():
|
778 |
+
"""Gestion des entrées YouTube"""
|
779 |
+
youtube_url = st.text_input("URL YouTube")
|
780 |
+
if youtube_url and st.button("Analyser"):
|
781 |
+
video_processor = VideoProcessor()
|
782 |
+
video_id = video_processor.extract_video_id(youtube_url)
|
783 |
+
|
784 |
+
if video_id:
|
785 |
+
st.video(youtube_url)
|
786 |
+
with st.spinner("Traitement de la vidéo..."):
|
787 |
+
transcription = video_processor.get_youtube_transcription(video_id)
|
788 |
+
if transcription:
|
789 |
+
process_and_display_results(None, None, transcription)
|
790 |
+
else:
|
791 |
+
audio_path = video_processor.download_youtube_audio(youtube_url)
|
792 |
+
process_and_display_results(audio_path)
|
793 |
+
|
794 |
+
def process_and_display_results(file_path=None, audio_bytes=None, transcription=None):
|
795 |
+
"""Traitement et affichage des résultats"""
|
796 |
+
try:
|
797 |
+
if transcription is None:
|
798 |
+
transcription = get_transcription(file_path, audio_bytes)
|
799 |
+
|
800 |
+
if transcription:
|
801 |
+
display_transcription_and_summary(transcription)
|
802 |
+
finally:
|
803 |
+
cleanup_temporary_files()
|
804 |
+
|
805 |
+
def get_transcription(file_path=None, audio_bytes=None) -> str:
|
806 |
+
"""Obtention de la transcription"""
|
807 |
+
if file_path:
|
808 |
+
path = file_path if isinstance(file_path, str) else save_uploaded_file(file_path)
|
809 |
+
elif audio_bytes:
|
810 |
+
path = save_audio_bytes(audio_bytes)
|
811 |
+
else:
|
812 |
+
return None
|
813 |
+
|
814 |
+
chunks = st.session_state.audio_processor.chunk_audio(path)
|
815 |
+
transcriptions = []
|
816 |
+
|
817 |
+
with st.expander("Transcription", expanded=False):
|
818 |
+
progress_bar = st.progress(0)
|
819 |
+
for i, chunk in enumerate(chunks):
|
820 |
+
transcription = st.session_state.audio_processor.transcribe_chunk(chunk)
|
821 |
+
if transcription:
|
822 |
+
transcriptions.append(transcription)
|
823 |
+
progress_bar.progress((i + 1) / len(chunks))
|
824 |
+
|
825 |
+
return " ".join(transcriptions) if transcriptions else None
|
826 |
+
|
827 |
+
def display_transcription_and_summary(transcription: str):
|
828 |
+
"""Affichage de la transcription et du résumé"""
|
829 |
+
st.subheader("Transcription")
|
830 |
+
st.text_area("Texte transcrit:", value=transcription, height=200)
|
831 |
+
|
832 |
+
st.subheader("Résumé et Analyse")
|
833 |
+
summary = get_summary(transcription)
|
834 |
+
st.markdown(summary)
|
835 |
+
|
836 |
+
# Génération et téléchargement des documents
|
837 |
+
generate_and_download_documents(summary)
|
838 |
+
|
839 |
+
# Option d'envoi par email
|
840 |
+
handle_email_sending(summary)
|
841 |
+
|
842 |
+
def get_summary(transcription: str) -> str:
|
843 |
+
"""Génération du résumé"""
|
844 |
+
chunks = st.session_state.audio_processor.split_text(transcription)
|
845 |
+
if len(chunks) > 1:
|
846 |
+
return st.session_state.audio_processor.summarize_long_transcription(transcription)
|
847 |
+
return st.session_state.audio_processor.generate_summary(transcription)
|
848 |
+
|
849 |
+
def generate_and_download_documents(summary: str):
|
850 |
+
"""Génération et téléchargement des documents"""
|
851 |
+
timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
|
852 |
+
|
853 |
+
# Génération PDF
|
854 |
+
pdf_filename = f"resume_{timestamp}.pdf"
|
855 |
+
pdf_path = PDFGenerator.create_pdf(summary, pdf_filename)
|
856 |
+
|
857 |
+
# Génération DOCX
|
858 |
+
docx_filename = f"resume_{timestamp}.docx"
|
859 |
+
docx_path = generate_docx(summary, docx_filename)
|
860 |
+
|
861 |
+
# Boutons de téléchargement
|
862 |
+
col1, col2 = st.columns(2)
|
863 |
+
with col1:
|
864 |
+
with open(pdf_path, "rb") as pdf_file:
|
865 |
+
st.download_button(
|
866 |
+
"📥 Télécharger PDF",
|
867 |
+
pdf_file,
|
868 |
+
file_name=pdf_filename,
|
869 |
+
mime="application/pdf"
|
870 |
+
)
|
871 |
+
|
872 |
+
with col2:
|
873 |
+
with open(docx_path, "rb") as docx_file:
|
874 |
+
st.download_button(
|
875 |
+
"📥 Télécharger DOCX",
|
876 |
+
docx_file,
|
877 |
+
file_name=docx_filename,
|
878 |
+
mime="application/vnd.openxmlformats-officedocument.wordprocessingml.document"
|
879 |
+
)
|
880 |
+
|
881 |
+
return pdf_path
|
882 |
+
|
883 |
+
def handle_email_sending(summary: str):
|
884 |
+
"""Gestion de l'envoi par email"""
|
885 |
+
st.subheader("📧 Recevoir le résumé par email")
|
886 |
+
recipient_email = st.text_input("Entrez votre adresse email:")
|
887 |
+
|
888 |
+
if st.button("Envoyer par email"):
|
889 |
+
if not is_valid_email(recipient_email):
|
890 |
+
st.error("Veuillez entrer une adresse email valide.")
|
891 |
+
return
|
892 |
+
|
893 |
+
with st.spinner("Envoi de l'email en cours..."):
|
894 |
+
pdf_path = generate_and_download_documents(summary)
|
895 |
+
email_sender = EmailSender(Config.SENDER_EMAIL, Config.SENDER_PASSWORD)
|
896 |
+
|
897 |
+
if email_sender.send_email(
|
898 |
+
recipient_email,
|
899 |
+
"Résumé de votre contenu audio/vidéo",
|
900 |
+
"Veuillez trouver ci-joint le résumé de votre contenu.",
|
901 |
+
pdf_path
|
902 |
+
):
|
903 |
+
st.success("Email envoyé avec succès!")
|
904 |
+
else:
|
905 |
+
st.error("Échec de l'envoi de l'email.")
|
906 |
+
|
907 |
+
|
908 |
+
def cleanup_temporary_files():
|
909 |
+
"""Nettoyage des fichiers temporaires"""
|
910 |
+
temp_files = ['temp_audio.mp3', 'temp_video.mp4']
|
911 |
+
for temp_file in temp_files:
|
912 |
+
if os.path.exists(temp_file):
|
913 |
+
try:
|
914 |
+
os.remove(temp_file)
|
915 |
+
except Exception:
|
916 |
+
pass
|
917 |
+
|
918 |
+
def process_document():
|
919 |
+
"""Traitement des documents"""
|
920 |
+
file = st.file_uploader("Chargez votre document", type=['pdf', 'docx', 'pptx', 'txt'])
|
921 |
+
password = st.text_input("Mot de passe (si protégé)", type="password")
|
922 |
+
|
923 |
+
if file:
|
924 |
+
try:
|
925 |
+
doc_processor = DocumentProcessor(
|
926 |
+
st.session_state.audio_processor.llm.model_name,
|
927 |
+
st.session_state.audio_processor.custom_prompt
|
928 |
+
)
|
929 |
+
text = process_document_with_password(file, password, doc_processor)
|
930 |
+
if text:
|
931 |
+
summary = doc_processor.summarize_text(text)
|
932 |
+
display_summary_and_downloads(summary)
|
933 |
+
except ValueError as e:
|
934 |
+
st.error(str(e))
|
935 |
+
|
936 |
+
def process_document_with_password(file, password: str, doc_processor: DocumentProcessor) -> Optional[str]:
|
937 |
+
"""Traitement des documents protégés par mot de passe"""
|
938 |
+
file_extension = os.path.splitext(file.name)[1].lower()
|
939 |
+
|
940 |
+
try:
|
941 |
+
if file_extension == '.pdf':
|
942 |
+
return doc_processor.process_protected_pdf(file, password)
|
943 |
+
elif file_extension in ['.docx', '.pptx']:
|
944 |
+
return doc_processor.process_protected_office(file, file_extension[1:], password)
|
945 |
+
elif file_extension == '.txt':
|
946 |
+
return file.read().decode('utf-8')
|
947 |
+
else:
|
948 |
+
st.error("Format de fichier non supporté")
|
949 |
+
return None
|
950 |
+
except ValueError as e:
|
951 |
+
st.error(str(e))
|
952 |
+
return None
|
953 |
+
|
954 |
+
def process_web():
|
955 |
+
"""Traitement des contenus web"""
|
956 |
+
url = st.text_input("URL du site web")
|
957 |
+
auth_required = st.checkbox("Authentification requise")
|
958 |
+
|
959 |
+
auth = None
|
960 |
+
if auth_required:
|
961 |
+
username = st.text_input("Nom d'utilisateur")
|
962 |
+
password = st.text_input("Mot de passe", type="password")
|
963 |
+
auth = {"username": username, "password": password}
|
964 |
+
|
965 |
+
if url and st.button("Analyser"):
|
966 |
+
try:
|
967 |
+
doc_processor = DocumentProcessor(
|
968 |
+
st.session_state.audio_processor.llm.model_name,
|
969 |
+
st.session_state.audio_processor.custom_prompt
|
970 |
+
)
|
971 |
+
text = doc_processor.scrape_web_content(url, auth)
|
972 |
+
if text:
|
973 |
+
summary = doc_processor.summarize_text(text)
|
974 |
+
display_summary_and_downloads(summary)
|
975 |
+
except ValueError as e:
|
976 |
+
st.error(str(e))
|
977 |
+
|
978 |
+
def display_summary_and_downloads(summary: str):
|
979 |
+
"""Affichage du résumé et options de téléchargement"""
|
980 |
+
st.markdown("### 📝 Résumé et Analyse")
|
981 |
+
st.markdown(summary)
|
982 |
+
|
983 |
+
timestamp = datetime.now().strftime('%Y%m%d_%H%M%S')
|
984 |
+
|
985 |
+
# Génération PDF
|
986 |
+
pdf_filename = f"resume_{timestamp}.pdf"
|
987 |
+
pdf_path = PDFGenerator.create_pdf(summary, pdf_filename)
|
988 |
+
|
989 |
+
# Génération DOCX
|
990 |
+
docx_filename = f"resume_{timestamp}.docx"
|
991 |
+
docx_path = generate_docx(summary, docx_filename)
|
992 |
+
|
993 |
+
# Boutons de téléchargement
|
994 |
+
col1, col2 = st.columns(2)
|
995 |
+
with col1:
|
996 |
+
with open(pdf_path, "rb") as pdf_file:
|
997 |
+
st.download_button(
|
998 |
+
"📥 Télécharger PDF",
|
999 |
+
pdf_file,
|
1000 |
+
file_name=pdf_filename,
|
1001 |
+
mime="application/pdf"
|
1002 |
+
)
|
1003 |
+
|
1004 |
+
with col2:
|
1005 |
+
with open(docx_path, "rb") as docx_file:
|
1006 |
+
st.download_button(
|
1007 |
+
"📥 Télécharger DOCX",
|
1008 |
+
docx_file,
|
1009 |
+
file_name=docx_filename,
|
1010 |
+
mime="application/vnd.openxmlformats-officedocument.wordprocessingml.document"
|
1011 |
+
)
|
1012 |
+
|
1013 |
+
# Option d'envoi par email
|
1014 |
+
st.markdown("### 📧 Recevoir le résumé par email")
|
1015 |
+
recipient_email = st.text_input("Entrez votre adresse email:")
|
1016 |
+
|
1017 |
+
if st.button("Envoyer par email"):
|
1018 |
+
if not is_valid_email(recipient_email):
|
1019 |
+
st.error("Veuillez entrer une adresse email valide.")
|
1020 |
+
else:
|
1021 |
+
with st.spinner("Envoi de l'email en cours..."):
|
1022 |
+
email_sender = EmailSender(Config.SENDER_EMAIL, Config.SENDER_PASSWORD)
|
1023 |
+
if email_sender.send_email(
|
1024 |
+
recipient_email,
|
1025 |
+
"Résumé de votre contenu",
|
1026 |
+
"Veuillez trouver ci-joint le résumé de votre contenu.",
|
1027 |
+
pdf_path
|
1028 |
+
):
|
1029 |
+
st.success("Email envoyé avec succès!")
|
1030 |
+
else:
|
1031 |
+
st.error("Échec de l'envoi de l'email.")
|
1032 |
+
|
1033 |
+
def save_audio_bytes(audio_bytes: bytes) -> str:
|
1034 |
+
"""Sauvegarde les bytes audio dans un fichier temporaire"""
|
1035 |
+
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
|
1036 |
+
file_path = f"recording_{timestamp}.wav"
|
1037 |
+
with open(file_path, 'wb') as f:
|
1038 |
+
f.write(audio_bytes)
|
1039 |
+
return file_path
|
1040 |
+
|
1041 |
+
if __name__ == "__main__":
|
1042 |
+
try:
|
1043 |
+
enhance_main()
|
1044 |
+
except Exception as e:
|
1045 |
+
st.error(f"Une erreur inattendue est survenue: {str(e)}")
|
1046 |
+
st.error("Veuillez réessayer ou contacter le support technique.")
|
1047 |
+
finally:
|
1048 |
+
cleanup_temporary_files()
|