Update tts.py

tts.py

@@ -1,7 +1,5 @@
-import torch
-from transformers import AutoTokenizer, AutoModelForTextToWaveform
+from transformers import pipeline
 import logging
-import numpy as np
 import soundfile as sf
 
 # Set up logging
@@ -9,13 +7,12 @@ logging.basicConfig(level=logging.DEBUG)
 
 MODEL_ID = "facebook/mms-tts-fao"
 
-# Try to load the model and tokenizer
+# Try to create the pipeline
 try:
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-    model = AutoModelForTextToWaveform.from_pretrained(MODEL_ID)
-    logging.info("Model and tokenizer loaded successfully.")
+    pipe = pipeline("text-to-speech", model=MODEL_ID)
+    logging.info("Pipeline created successfully.")
 except Exception as e:
-    logging.error(f"Error loading model or tokenizer: {e}")
+    logging.error(f"Error creating pipeline: {e}")
     raise
 
 def synthesize_speech(text):
@@ -25,24 +22,12 @@ def synthesize_speech(text):
             logging.error("Text input is empty.")
             return None
 
-        inputs = tokenizer(text, return_tensors="pt")
-        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-        model.to(device)
-        inputs = inputs.to(device)
+        result = pipe(text)
+        waveform = result["waveform"]
 
-        with torch.no_grad():
-            outputs = model.generate(**inputs)
-        
-        logging.info("Speech generated successfully.")
-
-        # Convert outputs to waveform
-        waveform = outputs.cpu().numpy().flatten()
-        # Normalize waveform to the range [-1, 1]
-        waveform = np.clip(waveform, -1.0, 1.0)
-        
-        # Convert waveform to audio format that Gradio can handle
+        # Save waveform to an audio file
         audio_path = "output.wav"
-        sf.write(audio_path, waveform, 16000)
+        sf.write(audio_path, waveform, 16000)  # Write the waveform using soundfile
         return audio_path
     except Exception as e:
         logging.error(f"Error during speech synthesis: {e}")