Fix generate_kwargs and sourced model

app.py

@@ -9,7 +9,7 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="WasuratS/whisper-base-danish-finetune-common-voice-11", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl")
@@ -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": "nl"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]
 
 
@@ -41,7 +41,7 @@ def speech_to_speech_translation(audio):
 
 title = "Cascaded STST - Danish to Dutch"
 description = """
-Demo for cascaded speech-to-speech translation (STST), mapping from source speech in Danish language to target speech in Dutch ! <br/> Demo uses my fine tuned OpenAI's [Whisper Base](WasuratS/whisper-base-danish-finetune-common-voice-11) model for speech translation, and my fine tuned Microsoft's
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in Danish language to target speech in Dutch ! <br/> Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and my fine tuned Microsoft's
 [SpeechT5 TTS](https://huggingface.co/WasuratS/speecht5_finetuned_voxpopuli_nl) 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")