Update app.py

app.py

@@ -5,10 +5,8 @@ import torch
 from transformers import VitsModel, AutoTokenizer, pipeline
 
 
-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")
 
 # load text-to-speech checkpoint
 model = VitsModel.from_pretrained("facebook/mms-tts-fra")
@@ -25,13 +23,11 @@ def translate(audio):
 
 
 def synthesise(text):
-    print(text)
     inputs = tokenizer(text=text, return_tensors="pt")
-    # inputs['input_ids'] = inputs['input_ids'].long()
     with torch.no_grad():
         speech = model(**inputs).waveform
     speech = speech[0]  # remove batch dimension
-    return speech.cpu()
+    return speech
 
 
 def speech_to_speech_translation(audio):