Update app.py

app.py

@@ -4,12 +4,14 @@ import torch
 from datasets import load_dataset
 
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+import spaces
 
 
-device = "cuda:0" if torch.cuda.is_available() else "cpu"
+# device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base")
+asr_pipe.to('cuda')
 
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
@@ -20,18 +22,18 @@ vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(devic
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
-
+@spaces.GPU
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["text"]
 
-
+@spaces.GPU
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")
     speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
     return speech.cpu()
 
-
+@spaces.GPU
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)