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"
 
@@ -12,10 +12,12 @@ 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("microsoft/speecht5_tts")
+# processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+# model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
 
-model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+model = VitsModel.from_pretrained("Matthijs/mms-tts-nld")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-nld")
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
@@ -25,11 +27,18 @@ def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]
 
-
 def synthesise(text):
-    inputs = processor(text=text, return_tensors="pt", padding='max_length', truncation=True)
-    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 synthesise(text):
+#     inputs = processor(text=text, return_tensors="pt", padding='max_length', truncation=True)
+#     speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+#     return speech.cpu()
 
 
 def speech_to_speech_translation(audio):