Update utils.py

utils.py

@@ -7,7 +7,7 @@ import os
 
 # News Extraction
 def fetch_news(company_name):
-    api_key = os.getenv("NEWSAPI_KEY", "14c0c7c5415b48e093fad78a1fd9581b")  # Use environment variable or replace with your key
+    api_key = os.getenv("NEWSAPI_KEY", "14c0c7c5415b48e093fad78a1fd9581b")
     url = f"https://newsapi.org/v2/everything?q={company_name}&apiKey={api_key}"
     response = requests.get(url)
     
@@ -32,7 +32,7 @@ def fetch_news(company_name):
 # Sentiment Analysis
 sentiment_analyzer = pipeline("sentiment-analysis")
 def analyze_sentiment(text):
-    result = sentiment_analyzer(text[:512])[0]  # Limit to 512 tokens
+    result = sentiment_analyzer(text[:512])[0]
     label = result['label']
     if label == "POSITIVE":
         return "Positive"
@@ -55,9 +55,20 @@ model = VitsModel.from_pretrained("facebook/mms-tts-hin")
 tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-hin")
 
 def generate_hindi_tts(text):
-    inputs = tokenizer(text, return_tensors="pt")
-    with torch.no_grad():
-        outputs = model(**inputs).waveform
-    audio_file = "output.wav"
-    wav.write(audio_file, rate=model.config.sampling_rate, data=outputs.squeeze().cpu().numpy())
-    return audio_file
+    try:
+        # Tokenize input text (limit length to avoid model issues)
+        inputs = tokenizer(text[:200], return_tensors="pt")  # Limit to 200 chars for stability
+        with torch.no_grad():
+            outputs = model(**inputs).waveform
+        
+        # Normalize audio data
+        audio_data = outputs.squeeze().cpu().numpy()
+        if audio_data.ndim > 1:
+            audio_data = audio_data[0]  # Ensure 1D array
+        
+        # Save as WAV file
+        audio_file = "output.wav"
+        wav.write(audio_file, rate=model.config.sampling_rate, data=audio_data)
+        return audio_file
+    except Exception as e:
+        raise Exception(f"TTS generation failed: {str(e)}")