Update app.py

app.py

@@ -3,7 +3,7 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, VitsModel, VitsTokenizer
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -11,11 +11,8 @@ device = "cuda:0" if torch.cuda.is_available() else "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("JFuellem/speecht5_finetuned_voxpopuli_de")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("JFuellem/speecht5_finetuned_voxpopuli_de").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+model = VitsModel.from_pretrained("Matthijs/mms-tts-deu")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-deu")
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
@@ -27,9 +24,12 @@ def translate(audio):
 
 
 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)
-    return speech.cpu()
+    inputs = tokenizer(text=text, return_tensors="pt")
+    
+    with torch.no_grad():
+        outputs = model(**inputs)
+    
+    return outputs.waveform[0]
 
 
 def speech_to_speech_translation(audio):