Update app.py

reverted to dutch

app.py

@@ -6,15 +6,15 @@ from datasets import load_dataset
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
 
 
-#device = "cuda:0" if torch.cuda.is_available() else "cpu"
-device = "cpu"
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+#device = "cpu"
 
 # load speech translation checkpoint
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
 #processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
-processor = SpeechT5Processor.from_pretrained("aoliveira/speecht5_finetuned_voxpopuli_sk")
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 model = SpeechT5ForTextToSpeech.from_pretrained("aoliveira/speecht5_finetuned_voxpopuli_sk").to(device)
 #model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
@@ -24,7 +24,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "sk"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]
 
 
@@ -43,8 +43,8 @@ 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 Slovak. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's 
-SpeechT5 TTS model fine tuned for the Slovak language for text-to-speech:
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Dutch. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's 
+SpeechT5 TTS model fine tuned for the Dutch language 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")
 """