Adicionado POC do serviço de audio (Adicionado Roberta)

app/api.py

@@ -41,7 +41,8 @@ def analyze(request: Request, data: dict):
     try:
         result = process_audio(
             video_base64=video_base64,
-            audio_base64=audio_base64
+            audio_base64=audio_base64,
+            logger=logger
         )
 
         log("Finalizado analise de audio", logger=logger)

app/audio_processor.py

@@ -3,6 +3,7 @@ import tempfile
 import os
 from app.config import WHISPER_MODEL, WHISPER_MODEL_PATH
 os.environ["XDG_CACHE_HOME"] = str(WHISPER_MODEL_PATH) # "./models"
+from app.logger import log
 from app.sentiment_model import analyze_sentiment
 import whisper
 from moviepy.editor import VideoFileClip
@@ -41,7 +42,7 @@ def extract_audio_from_video(video_path):
 # -------------------------
 # 🔥 PROCESSAMENTO PRINCIPAL
 # -------------------------
-def process_audio(video_base64=None, audio_base64=None):
+def process_audio(video_base64=None, audio_base64=None, logger=None):
 
     video_path = None
     audio_path = None
@@ -52,14 +53,14 @@ def process_audio(video_base64=None, audio_base64=None):
         # -------------------------
 
         if audio_base64:
-            print("Recebido áudio base64")
+            log("Recebido áudio base64", logger=logger)
             audio_path = save_base64_to_file(audio_base64, ".wav")
 
         elif video_base64:
-            print("Recebido vídeo base64")
+            log("Recebido vídeo base64", logger=logger)
             video_path = save_base64_to_file(video_base64, ".mp4")
 
-            print("Extraindo áudio do vídeo...")
+            log("Extraindo áudio do vídeo...", logger=logger)
             audio_path = extract_audio_from_video(video_path)
 
         else:
@@ -69,7 +70,7 @@ def process_audio(video_base64=None, audio_base64=None):
         # 🎯 2. WHISPER
         # -------------------------
 
-        print("Transcrevendo com Whisper...")
+        log("Transcrevendo com Whisper...", logger=logger)
 
         result = whisper_model.transcribe(
             audio_path,
@@ -78,15 +79,13 @@ def process_audio(video_base64=None, audio_base64=None):
             fp16=False  # importante para CPU
         )
 
-        # text = result["text"]
-
         text = " ".join([seg["text"] for seg in result["segments"]])
 
         # -------------------------
         # 🎯 3. ROBERTA (cardiffnlp/twitter-xlm-roberta)
         # -------------------------
 
-        sentiment, score = analyze_sentiment(text)
+        sentiment, score = analyze_sentiment(text, logger)
 
         print(f"Sentimento: {sentiment} ({score})")
 

app/sentiment_model.py

@@ -1,3 +1,4 @@
+from app.logger import log
 from transformers import pipeline
 
 _pipeline = None
@@ -30,26 +31,7 @@ def split_text(text, max_length=300):
 
     return chunks
 
-def analyze_sentiment_v1(text: str):
-    model = get_pipeline()
-
-    #result = model(text[:512])[0]
-
-    result = model(text)[0]
-
-    label = result["label"]
-    score = result["score"]
-
-    if label == "LABEL_0":
-        sentiment = "negative"
-    elif label == "LABEL_1":
-        sentiment = "neutral"
-    else:
-        sentiment = "positive"
-
-    return sentiment, score
-
-def analyze_sentiment(text: str):
+def analyze_sentiment(text: str, logger=None):
     model = get_pipeline()
 
     # 🔥 quebra em partes menores
@@ -58,6 +40,9 @@ def analyze_sentiment(text: str):
     results = []
 
     for chunk in chunks:
+        if not chunk.strip():
+            continue
+
         r = model(chunk)[0]
         results.append(r)
 
@@ -72,11 +57,19 @@ def analyze_sentiment(text: str):
     }
 
     if not results:
-        return "neutral...", 0
+        log("Não houve resultado", "warning", logger=logger)
+        return "neutral", 0
 
     total = 0
     for r in results:
-        total += score_map[r["label"]] * r["score"]
+        label = r["label"]
+        score = r["score"]
+
+        if label not in score_map:
+            continue
+
+        log(f"Label: {label} | Score: {score}")
+        total += score_map[label] * score
 
     avg = total / len(results)