Update app.py

app.py

@@ -4,7 +4,7 @@ import torch
 from datasets import load_dataset
 
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-from transformers import VitsModel, VitsTokenizer
+
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
@@ -12,20 +12,17 @@ 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("Matthijs/mms-tts-deu")
-tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-deu")
+processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+model = SpeechT5ForTextToSpeech.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 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": "transcribe", "language": "de"})
-    return outputs["text"]
-
-
-#
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
+    return text
+    
 def synthesise(text):
     inputs = tokenizer(text, return_tensors="pt")
     with torch.no_grad():
@@ -42,7 +39,7 @@ 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 OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and my fine tuned Microsoft's
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any X 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")