gröbste Telegram-Fehler beseitigt, aiornot-Library

main_tg.py

@@ -63,15 +63,13 @@ if __name__ == "__main__":
         if 'aiornot_ai_score' not in post: 
             if post['video'] is not None:
                 # Audio des Videos analysieren
-                post['aiornot_ai_score'] = aiornot_wrapper(post['video'].get('url'), is_image = False)
+                post['aiornot_ai_score'] = aiornot_wrapper(post['video'].get('file'), is_image = False)
                 print("Video: AIORNOT-Score")
                 # Bild analysieren
                 # Das hier ist für die Galerie: AIORNOT kann derzeit
                 # keine base64-Strings checken. 
-                # Das Problem an den URLs der Photos ist: sie sind nicht garantiert. 
-                base64_image = post['photo'].get('image',None) 
-                image = f"data:image/jpeg;base64, {base64_image}"
-            post['aiornot_ai_score'] = aiornot_wrapper(post['photo'].get('url'))
+            elif post['photo'] is not None: 
+                post['aiornot_ai_score'] = aiornot_wrapper(post['photo'].get('file'))
             print("AIORNOT-AI-Score: {post['aiornot_ai_score']}")
             # Videos kann man nur über das Audio auf KI checken. 
             # Muss ich erst noch implementieren. 

requirements.txt

+aiornot==0.0.5
+beautifulsoup4==4.12.3
+numpy==2.0.2
+ollama==0.4.4
+openai==1.58.1
+openai-whisper==20240930
+openpyxl==3.1.5
+pandas==2.2.3
+pathlib==1.0.1
+pydub==0.25.1
+replicate==1.0.4
+soundfile==0.13.0
+toml==0.10.2
\ No newline at end of file

src/aichecker/aiornot.py

@@ -62,7 +62,7 @@ def query_aiornot(content, is_image = False):
             return None
     # Dateiname? Dann mit Multipart-Header 
     if not (content.startswith("http://") or content.startswith("https://")):
-        fname = 
+        fname = content
         headers = {
             'Authorization': f"Bearer {api_key}",
             'Accept': 'application/json',

src/aichecker/check_tg.py

@@ -108,7 +108,7 @@ def tgc_clean(cname):
 def save_url(fname, name, mdir="./media"):
     # Die Medien-URLs bekommen oft einen Parameter mit übergeben; deswegen nicht nur
     # "irgendwas.ogg" berücksichtigen, sondern auch "irgendwas.mp4?nochirgendwas"
-    content_ext = re.search("\.[a-zA-Z0-9]+(?=\?|$)",fname).group(0)
+    content_ext = re.search(r"\.[a-zA-Z0-9]+(?=\?|$)",fname).group(0)
     content_file = f"{mdir}/{name}{content_ext}"
     try:
         os.makedirs(os.path.dirname(content_file), exist_ok=True)
@@ -186,8 +186,12 @@ def tg_post_parse(b, save = True, describe = True):
     # Sprachnachricht tgme_widget_message_voice https://t.me/fragunsdochDasOriginal/27176
     if b.select_one('audio.tgme_widget_message_voice') is not None:
         # Link auf OGG-Datei
-        voice_url = b.select_one('audio.tgme_widget_message_voice')[url]
+        voice_url = b.select_one('audio.tgme_widget_message_voice')['src']
         voice_duration = b.select_one('time.tgme_widget_message_voice_duration').get_text()
+        voice = {
+            'url': voice_url,
+            'duration': voice_duration,
+        }
         # Für Transkription immer lokale Kopie anlegen
         if save or describe: 
             voice['file'] = save_url(voice_url, f"{channel}_{b_nr}_voice")
@@ -321,7 +325,7 @@ def tgc_read_range(cname, n1=1, n2=None, save=True, describe = True):
     # Zuerst: Nummer des letzten Posts holen
     profile = tgc_profile(cname)
     # Sicherheitscheck: erste Post-Nummer überhaupt schon gepostet?
-    max_nr = profile['n_post']
+    max_nr = profile['n_posts']
     if n1 > max_nr: 
         return None
     loop = True
@@ -388,12 +392,7 @@ def check_tg_list(posts, check_images = True):
                 post['aiornot_ai_score'] = aiornot_wrapper(post['video'].get('file'), is_image = False)
             elif post['photo'] is not None:
                 # Bild analysieren
-                # Das hier ist für die Galerie: AIORNOT kann derzeit
-                # keine base64-Strings checken. 
-                # Das Problem an den URLs der Photos ist: sie sind nicht garantiert. 
-                base64_image = post['photo'].get('image',None) 
-                image = f"data:image/jpeg;base64, {base64_image}"
-                post['aiornot_ai_score'] = aiornot_wrapper(post['photo'].get('url'))
+                post['aiornot_ai_score'] = aiornot_wrapper(post['photo'].get('file'), is_image = True)
     return posts
 # Wrapper für die check_tg_list Routine. 
 # Gibt Resultate als df zurück, arbeitet aber hinter den Kulissen mit 

src/aichecker/check_wrappers.py

 from .detectora import query_detectora
-from .aiornot import query_aiornot
+# from .aiornot import query_aiornot
 from .transcribe import gpt4_description
 
+# Alternative zu meinen selbst geschriebenen aiornot-Routinen: 
+# https://github.com/aiornotinc/aiornot-python
+# Installieren mit
+#    pip install aiornot
+
+from aiornot import Client
+
 # Konstante 
 d_thresh = .8 # 80 Prozent 
 limit = 25 # Posts für den Check
@@ -20,20 +27,25 @@ def detectora_wrapper(text: str):
     return score
 
 def aiornot_wrapper(content, is_image = True):
+    aiornot_client = Client()
     # Verpackung. Fortschrittsbalken.
     if content is None:
         print(" ", end="")
         return
     # Fortschrittsbalken
     print("?", end="")
-    report = query_aiornot(content, is_image)
+    is_url = (content.startswith("http://") or content.startswith("https://"))
+    if is_image:
+        response = aiornot_client.image_report_by_url(content) if is_url else aiornot_client.image_report_by_file(content)
+    else: 
+        response = aiornot_client.audio_report_by_url(content) if is_url else aiornot_client.audio_report_by_file(content)        
     # Beschreibung: https://docs.aiornot.com/#5b3de85d-d3eb-4ad1-a191-54988f56d978   
-    if report is not None:  
+    if response is not None:  
         aiornot_dict = ({
-            'aiornot_score': report['verdict'],
+            'aiornot_score': response.report.verdict,
             # Unterscheidung: Bilder haben den Confidence score im Unter-Key 'ai'
-            'aiornot_confidence': report['ai']['confidence'] if 'ai' in report else report['confidence'],
-            'aiornot_generator': report['generator'] if 'generator' in report else 'Audio',
+            'aiornot_confidence': response.report.ai.confidence if hasattr(response.report, 'ai') else response.report.confidence,
+            'aiornot_generator': response.report.generator if hasattr(response.report, 'generator') else None,
         })
         print(f"\b{'X' if aiornot_dict['aiornot_score'] != 'human' else '.'}",end="")
         return aiornot_dict
@@ -50,9 +62,9 @@ def bsky_aiornot_wrapper(did,embed):
     if 'images' in embed:
         images = embed['images']
         desc = []
-        for i in images:
+        for image in images:
             # Construct an URL for the image thumbnail (normalised size)
-            link = i['image']['ref']['$link']
+            link = image['image']['ref']['$link']
             i_url = f"https://cdn.bsky.app/img/feed_thumbnail/plain/{did}/{link}"
             aiornot_report = aiornot_wrapper(i_url)
             aiornot_report['gpt4_description'] = gpt4_description(image)

src/aichecker/transcribe.py

@@ -66,7 +66,7 @@ def ai_description(image):
     # Return ai-generated description
     return desc2
 
-def transcribe(fname):
+def transcribe(fname, use_api = False):
     # Wrapper; ruft eine der drei Whisper-Transcribe-Varianten auf. 
     # Favorit: das beschleunigte whisper-s2t
     # (das aber erst CTranslate2 mit METAL-Unterstützung braucht auf dem Mac
@@ -76,8 +76,10 @@ def transcribe(fname):
     if ".ogg" in fname.lower():
         fname = convert_ogg_to_m4a(fname)
     try: 
-        text = transcribe_whisper(fname)
-        #text = transcribe_api(fname)
+        if use_api:
+            text = transcribe_api(fname)
+        else: 
+            text = transcribe_whisper(fname)
         # return transcribe_jax(audio)
         # return transcribe_ws2t(audio)
         return text