Update app.py

app.py

@@ -14,10 +14,10 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
-#model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-#vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
-model = VitsModel.from_pretrained("facebook/mms-tts-fra")
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-fra")
+model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+#model = VitsModel.from_pretrained("facebook/mms-tts-fra")
+#tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-fra")
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
@@ -29,17 +29,19 @@ def translate(audio):
 
 
 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)
-    inputs = tokenizer(text, return_tensors="pt")
-    speech = model(inputs["input_ids"])
-    return speech["waveform"]
+    inputs = processor(text=text, return_tensors="pt")
+    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    #inputs = tokenizer(text, return_tensors="pt")
+    #speech = model(inputs["input_ids"])
+    #output = speech["waveform"]
+    return speech
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
-    synthesised_speech = (synthesised_speech.detach().numpy() * 32767).astype(np.int16)
+    #synthesised_speech = (synthesised_speech.detach().numpy() * 32767).astype(np.int16)
+    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech