changed language to Vietnamese

app.py

@@ -12,15 +12,15 @@ 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")
-tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts")
+model = VitsModel.from_pretrained("facebook/mms-tts-vie")
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-vie")
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "lt"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "vi"})
     return outputs["text"]
 
 
@@ -40,7 +40,7 @@ def speech_to_speech_translation(audio):
 
 title = "Cascaded STST"
 description = """
-Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Lithuanian.
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Vietnamese.
 Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Meta's [MMS TTS](https://huggingface.co/facebook/mms-tts-deu) model for text-to-speech:
 
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")