.

.gitignore

@@ -0,0 +1 @@
+*.mbp

base-config.yaml

@@ -14,5 +14,8 @@ allowed_rooms: []
 # Optional text to guide the model's style or to continue from a previous audio segment. The prompt should match the language of the audio.
 prompt: "以下是普通话录制的会议记录："
 
+#Optional should OpenAI re-read the transcript to detect reminders and events ?
+search_reminders_and_events: false
+
 # The language of the input audio. Providing the input language in ISO-639-1 format (Chinese: "zh") will improve accuracy and latency, leave empty by default
 language:

maubot.yaml

@@ -1,6 +1,6 @@
 maubot: 0.1.0
 id: nigzu.com.maubot-stt
-version: 0.2.6
+version: 0.3.4
 license: MIT
 modules:
 - openai-whisper

openai-whisper.py

@@ -9,9 +9,15 @@ from mautrix.client import Client
 from maubot.handlers import event
 from maubot import Plugin, MessageEvent
 from mautrix.errors import MatrixRequestError
-from mautrix.types import EventType, MessageType, RelationType, TextMessageEventContent, Format,RelatesTo,InReplyTo
+from mautrix.types import (
+    EventType, MessageType, RelationType,
+    TextMessageEventContent, Format, RelatesTo, InReplyTo
+)
 from mautrix.util.config import BaseProxyConfig, ConfigUpdateHelper
 
+ALLOWED_EXTENSIONS = ['flac', 'm4a', 'mp3', 'mp4', 'mpeg', 'mpga', 'oga', 'ogg', 'wav', 'webm']
+ALLOWED_MIME_TYPES = ['audio/flac', 'audio/mp4', 'video/mpeg', 'audio/ogg', 'audio/wav', 'video/webm']
+
 class Config(BaseProxyConfig):
     def do_update(self, helper: ConfigUpdateHelper) -> None:
         helper.copy("whisper_endpoint")
@@ -19,6 +25,7 @@ class Config(BaseProxyConfig):
         helper.copy("allowed_users")
         helper.copy("allowed_rooms")
         helper.copy("prompt")
+        helper.copy("search_reminders_and_events")
         helper.copy("language")
 
 class WhisperPlugin(Plugin):
@@ -32,6 +39,7 @@ class WhisperPlugin(Plugin):
         self.language = self.config['language']
         self.allowed_users = self.config['allowed_users']
         self.allowed_rooms = self.config['allowed_rooms']
+        self.search_reminders_and_events = self.config['search_reminders_and_events']
         self.log.debug("Whisper plugin started")
 
     async def should_respond(self, event: MessageEvent) -> bool:
@@ -44,7 +52,14 @@ class WhisperPlugin(Plugin):
         if self.allowed_rooms and event.room_id not in self.allowed_rooms:
             return False
 
-        return event.content.msgtype == MessageType.AUDIO or event.content.msgtype == MessageType.FILE
+        # Extraction de la partie principale du MIME type (avant les éventuels paramètres)
+        mime_type = ""
+        if event.content.info and event.content.info.mimetype:
+            mime_type = event.content.info.mimetype.split(";")[0]
+        if mime_type not in ALLOWED_MIME_TYPES:
+            return False
+
+        return event.content.msgtype in (MessageType.AUDIO, MessageType.FILE)
 
     @event.on(EventType.ROOM_MESSAGE)
     async def on_message(self, event: MessageEvent) -> None:
@@ -56,10 +71,24 @@ class WhisperPlugin(Plugin):
             await self.client.set_typing(event.room_id, timeout=99999)
 
             audio_bytes = await self.client.download_media(url=event.content.url)
-            transcription = await self.transcribe_audio(audio_bytes)
+            if not audio_bytes:
+                await event.respond("Erreur lors du téléchargement du fichier audio.")
+                return
+
+            # Récupère le nom de fichier s'il est défini, sinon utilise une valeur par défaut
+            filename = getattr(event.content, "filename", "audio.mp3")
+            # Utilise le MIME type tel quel, ou une valeur par défaut
+            mime_type = event.content.info.mimetype if event.content.info and event.content.info.mimetype else "audio/mpeg"
+
+            transcription = await self.transcribe_audio(audio_bytes, filename, mime_type)
+            
+            # Si l'étude est activée, on utilise son résultat uniquement si celui-ci renvoie une commande
+            if self.search_reminders_and_events:
+                studied = await self.study_transcribe(transcription)
+                if studied.startswith("!rappel") or studied.startswith("!agenda"):
+                    transcription = studied
 
             await self.client.set_typing(event.room_id, timeout=0)
-
             content = TextMessageEventContent(
                 msgtype=MessageType.TEXT,
                 body=transcription,
@@ -79,31 +108,65 @@ class WhisperPlugin(Plugin):
             self.log.exception(f"Something went wrong: {e}")
             await event.respond(f"Something went wrong: {e}")
 
-
-    async def transcribe_audio(self, audio_bytes: bytes) -> str:
+    async def transcribe_audio(self, audio_bytes: bytes, filename: str = "audio.mp3", mime_type: str = "audio/mpeg") -> str:
         headers = {
             "Authorization": f"Bearer {self.api_key}"
         }
         data = aiohttp.FormData()
-        data.add_field('file', audio_bytes, filename='audio.mp3', content_type='audio/mpeg')
+        data.add_field('file', audio_bytes, filename=filename, content_type=mime_type)
         data.add_field('model', 'whisper-1')
         if self.prompt:
-            data.add_field('prompt', f"{self.prompt}")
+            data.add_field('prompt', self.prompt)
         if self.language:
-            data.add_field('language', f"{self.language}")
+            data.add_field('language', self.language)
 
         try:
             async with aiohttp.ClientSession() as session:
                 async with session.post(self.whisper_endpoint, headers=headers, data=data) as response:
                     if response.status != 200:
-                        self.log.error(f"Error response from API: {await response.text()}")
-                        return f"Error: {await response.text()}"
+                        error_text = await response.text()
+                        self.log.error(f"Error response from API: {error_text}")
+                        return f"Error: {error_text}"
                     response_json = await response.json()
                     return response_json.get("text", "Sorry, I can't transcribe the audio.")
         except Exception as e:
             self.log.exception(f"Failed to transcribe audio, msg: {e}")
             return "Sorry, an error occurred while transcribing the audio."
 
+    async def study_transcribe(self, transcription: str) -> str:
+        prompt = f"""
+        Voici la transcription du message vocal :
+
+        {transcription}
+        Ton objectif est d'analyser cette transcription afin de déterminer si l'utilisateur tente de créer un rappel ou un évènement.
+
+        - Si l'utilisateur essaie de créer un rappel, la sortie doit prendre la forme :
+        !rappel <date> <message>
+        - Si l'utilisateur essaie de créer un évènement, la sortie doit prendre la forme :
+        !agenda ##ROOM## <message>
+        - Si l'utilisateur ne cherche ni à créer un rappel ni un évènement, renvoie seulement la transcription telle quelle, sans ajout d'explication, de texte supplémentaire ou de ponctuation superflue.
+        Ne fournis aucun autre texte ni explication dans ta réponse, uniquement la sortie demandée.
+        """
+
+        url = "https://api.openai.com/v1/chat/completions"
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.api_key}"
+        }
+
+        data = {
+            "model": "gpt-4",
+            "messages": [
+                {"role": "user", "content": prompt}
+            ],
+            "temperature": 0.7
+        }
+
+        async with aiohttp.ClientSession() as session:
+            async with session.post(url, headers=headers, json=data) as response:
+                response_json = await response.json()
+                return response_json.get('choices', [])[0].get('message', {}).get('content', transcription)
+
     @classmethod
     def get_config_class(cls) -> Type[BaseProxyConfig]:
         return Config