changed model and tokeniser to general MMS TTS; updated synthesise()

app.py

@@ -12,8 +12,8 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-model = VitsModel.from_pretrained("facebook/mms-tts-deu")
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-deu")
+model = VitsModel.from_pretrained("facebook/mms-tts")
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts")
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
@@ -26,11 +26,9 @@ def translate(audio):
 
 def synthesise(text):
     inputs = tokenizer(text, return_tensors="pt")
-    input_ids = inputs["input_ids"]
     with torch.no_grad():
-        outputs = model(input_ids)
-    speech = outputs["waveform"]
-    return speech.cpu()
+        output = model(**inputs).waveform
+    return output.cpu()
 
 
 def speech_to_speech_translation(audio):