Módulo de audio y avance imágenes (schemas y openAI)

app/services/audio/transcription_adapters.py

@@ -10,9 +10,14 @@ Propósito:
 
 import tempfile
 import os
+import json
+from dotenv import load_dotenv
 from fastapi import HTTPException 
 from openai import OpenAI, AsyncOpenAI
 import assemblyai as aai
+from deepgram import (
+    DeepgramClient,
+)
 from app.core.config import settings
 from app.schemas.audio_standard import AudioRequestFile
 from app.schemas.audio_standard import StandardTranscriptionResult
@@ -24,6 +29,7 @@ async def transcribe_audio_with_provider(audio_request: AudioRequestFile) -> Sta
     """
     Función de adaptador para transcribir audio usando el proveedor de IA configurado.
     """
+    load_dotenv()
     provider = audio_request.provider.lower()
     
     match provider:
@@ -31,6 +37,8 @@ async def transcribe_audio_with_provider(audio_request: AudioRequestFile) -> Sta
             return await transcribe_with_openai(audio_request)
         case "assemblyai":
             return await transcribe_with_assemblyai(audio_request)
+        case "deepgram":
+            return await transcribe_with_deepgram(audio_request)
         case _:
             raise ValueError(f"Proveedor de IA no soportado: {audio_request.provider}")
 
@@ -39,6 +47,7 @@ async def transcribe_with_openai(audio_request: AudioRequestFile) -> StandardTra
     """
     Función de adaptador para transcribir audio usando OpenAI.
     """
+    
     client = AsyncOpenAI(api_key=settings.OPENAI_API_KEY)
     
     audio_content = await audio_request.file.read()
@@ -112,7 +121,8 @@ async def transcribe_with_assemblyai(audio_request: AudioRequestFile) -> Standar
             temp_audio_path = temp_audio.name
         
         #Definimos el modelo a usar
-        config  = aai.TranscriptionConfig(language_code="es", speaker_labels=audio_request.diarization, 
+        config  = aai.TranscriptionConfig(speech_models = [audio_request.model], 
+                                          language_code="es", speaker_labels=audio_request.diarization, 
                                           sentiment_analysis=audio_request.sentiment)
         
         transcription_obj = aai.Transcriber(config=config).transcribe(temp_audio_path)
@@ -159,5 +169,71 @@ async def transcribe_with_assemblyai(audio_request: AudioRequestFile) -> Standar
                 os.unlink(temp_audio_path)
             except Exception:
                 pass   
-    # Aquí iría la implementación específica para AssemblyAI, similar a la de OpenAI pero usando su SDK y formato de respuesta
-    pass
+
+async def transcribe_with_deepgram(audio_request: AudioRequestFile):
+    """
+    Función de adaptador para transcribir audio usando Deepgram.
+    """
+    
+    #Inicializamos el cliente de Deepgram
+    deepgram = DeepgramClient(api_key=settings.DEEPGRAM_API_KEY)
+    audio_content = await audio_request.file.read()
+    temp_audio_path = None 
+    
+    # Validar el audio antes de continuar
+    validate_audio_request(audio_request, audio_content)
+    
+    try:
+        # Crear archivo temporal para el audio
+        with tempfile.NamedTemporaryFile(
+            delete=False,
+            suffix=os.path.splitext(audio_request.file.filename)[1]
+        ) as temp_audio:
+            temp_audio.write(audio_content)
+            temp_audio_path = temp_audio.name
+        with open(temp_audio_path, "rb") as audio_file:
+            response = deepgram.listen.v1.media.transcribe_file(
+                request=audio_file.read(),
+                model=audio_request.model,
+                sentiment=audio_request.sentiment,
+                utterances=audio_request.diarization,
+                diarize=audio_request.diarization,
+                # Deepgram no tiene una opción específica de "timestamps", pero sí devuelve marcas de tiempo por segmento, así que no es necesario un parámetro adicional para eso
+                smart_format=True,
+                language='es',
+            )
+            response_json = json.loads(response.json())
+                        
+            result = StandardTranscriptionResult(
+                status="success",
+                original_filename=audio_request.file.filename,
+                full_transcript=response_json.get("results", {}).get("channels", [{}])[0].get("alternatives", [{}])[0].get("transcript", ""),
+                model_used=audio_request.model,
+                provider_used="Deepgram",
+                confidence_score=response_json.get("results", {}).get("channels", [{}])[0].get("alternatives", [{}])[0].get("confidence"),
+                segments=[
+                    {
+                        "text": sentence.get("text", ""),
+                        "speaker": f"Speaker {paragraph.get('speaker')}" if audio_request.diarization and paragraph.get("speaker") is not None else None,
+                        "start_time": sentence.get("start") if audio_request.timestamps else None,
+                        "end_time": sentence.get("end") if audio_request.timestamps else None,
+                        "sentiment": sentence.get("sentiment") if audio_request.sentiment else None 
+                    }
+                    for paragraph in response_json.get("results", {}).get("channels", [{}])[0].get("alternatives", [{}])[0].get("paragraphs", {}).get("paragraphs", [])
+                    for sentence in paragraph.get("sentences", [])
+                ] if (audio_request.diarization or audio_request.timestamps or audio_request.sentiment) else None
+            )
+            return result
+    except Exception as e:
+        # Capturamos cualquier error de OpenAI o de lectura de archivos
+        raise HTTPException(
+            status_code=500,
+            detail=f"Error transcribiendo el audio: {str(e)}"
+        )
+        
+    finally:
+        if temp_audio_path and os.path.exists(temp_audio_path):
+            try:
+                os.unlink(temp_audio_path)
+            except Exception:
+                pass