Update app.py

app.py

@@ -22,7 +22,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "hu"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "hu"})
     return outputs["text"]
 
 
@@ -41,7 +41,7 @@ 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 English.
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Hungarian.
 Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation,
 and a [finetuned version](https://huggingface.co/catactivationsound/speecht5_finetuned_voxpopuli_hu)
 of Microsoft's [SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) for text-to-speech: