Update app.py

app.py

@@ -28,23 +28,21 @@ def tts_generate(text):
         inputs = processor(text=text, return_tensors="pt").to(device)
         print("✅ Text processed.")
 
-        # Generate mel spectrogram
-        print("🎤 Generating speech...")
+        # Generate waveform directly (with vocoder)
+        print("🎤 Generating speech waveform...")
         with torch.no_grad():
-            mel = model.generate_speech(inputs["input_ids"], speaker_embedding)
-        print("✅ Mel spectrogram generated.")
-
-        # Convert mel spectrogram to waveform
-        print("🎚️ Vocoding waveform...")
-        waveform = vocoder(mel)
-        waveform = waveform.cpu()
+            waveform = model.generate_speech(
+                inputs["input_ids"],
+                speaker_embedding,
+                vocoder=vocoder
+            )
         print("✅ Waveform generated.")
 
         # Save waveform
         output_path = "output.wav"
         if waveform.dim() == 1:
-            waveform = waveform.unsqueeze(0) 
-        torchaudio.save(output_path, waveform, sample_rate=16000)
+            waveform = waveform.unsqueeze(0)
+        torchaudio.save(output_path, waveform.cpu(), sample_rate=16000)
         print(f"💾 Audio saved to {output_path}")
 
         return output_path