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"
 
@@ -12,9 +12,8 @@ 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
-processor = SpeechT5Processor.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl").to(device)
+model = VitsModel.from_pretrained("Matthijs/mms-tts-deu")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-deu")
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
@@ -22,23 +21,25 @@ 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": "nl"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "de"})
     return outputs["text"]
 
 
+#
 def synthesise(text):
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    inputs = tokenizer(text, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model(inputs["input_ids"])
+    speech = outputs.audio[0]
+    
     return speech.cpu()
 
-
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech
 
-
 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