Update app.py

app.py

@@ -77,15 +77,11 @@ def generateAudio(text_to_audio, s3_save_as, key_id):
         # Preprocess text and recortar
         text = cut_text(text, max_tokens=500)
 
-        # Divide el texto en segmentos de 30 palabras
+        # Verificar si el texto tiene menos de 30 palabras
         palabras = text.split()
-        segmentos = [' '.join(palabras[i:i+30])
-                     for i in range(0, len(palabras), 30)]
-
-        # Generar audio para cada segmento y combinarlos
-        audio_segments = []
-        for segment in segmentos:
-            inputs = processor(text=segment, return_tensors="pt").to(device)
+        if len(palabras) <= 30:
+            # Generar audio para el texto completo
+            inputs = processor(text=text, return_tensors="pt").to(device)
             if speaker is not None:
                 speaker_embeddings = torch.tensor(
                     embeddings_dataset[speaker]["xvector"]).unsqueeze(0).to(device)
@@ -93,18 +89,42 @@ def generateAudio(text_to_audio, s3_save_as, key_id):
                 speaker_embeddings = torch.randn((1, 512)).to(device)
             speech = model.generate_speech(
                 inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
-            audio_segments.append(speech)
-
-        combined_audio = torch.cat(audio_segments, dim=0)
+            combined_audio = speech
+        else:
+            # Divide el texto en segmentos de 30 palabras
+            segmentos = [' '.join(palabras[i:i+30])
+                         for i in range(0, len(palabras), 30)]
+
+            # Generar audio para cada segmento y combinarlos
+            audio_segments = []
+            for segment in segmentos:
+                inputs = processor(
+                    text=segment, return_tensors="pt").to(device)
+                if speaker is not None:
+                    speaker_embeddings = torch.tensor(
+                        embeddings_dataset[speaker]["xvector"]).unsqueeze(0).to(device)
+                else:
+                    speaker_embeddings = torch.randn((1, 512)).to(device)
+                speech = model.generate_speech(
+                    inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
+                audio_segments.append(speech)
+
+            if len(audio_segments) > 0:
+                combined_audio = torch.cat(audio_segments, dim=0)
+            else:
+                combined_audio = None
 
-        # Crear objeto BytesIO para almacenar el audio
-        audio_buffer = BytesIO()
-        sf.write(audio_buffer, combined_audio.cpu().numpy(),
-                 samplerate=16000, format='WAV')
-        audio_buffer.seek(0)
+        if combined_audio is not None:
+            # Crear objeto BytesIO para almacenar el audio
+            audio_buffer = BytesIO()
+            sf.write(audio_buffer, combined_audio.cpu().numpy(),
+                     samplerate=16000, format='WAV')
+            audio_buffer.seek(0)
 
-        # Guardar el audio combinado en S3
-        save_audio_to_s3(audio_buffer)
+            # Guardar el audio combinado en S3
+            save_audio_to_s3(audio_buffer)
+        else:
+            print("File with content null")
 
     save_text_to_speech(text_to_audio, 2271)
     return s3_save_as
@@ -167,6 +187,7 @@ def list_s3_files():
 
 
 demo = gr.Blocks()
+
 with demo:
 
     text = gr.Textbox()