Update app.py

app.py

@@ -4,6 +4,8 @@ import torch
 from datasets import load_dataset
 import librosa
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import VitsModel, VitsTokenizer
+    
 
 target_dtype = np.int16
 max_range = np.iinfo(target_dtype).max
@@ -23,19 +25,28 @@ vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(devic
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
+model_mms = VitsModel.from_pretrained("facebook/mms-tts-nld")
+tokenizer_mms = VitsTokenizer.from_pretrained("facebook/mms-tts-nld")
+
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"language": "nl","task": "transcribe"})
     return outputs["text"]
 
 
-def synthesise(text):
+def synthesise_speechT5(text):
     inputs = processor(text=text, padding='max_length', truncation=True,max_length=600,return_tensors="pt")
     print(inputs)
     speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device),vocoder=vocoder)
     return speech.cpu()
 
-
+def synthesise(text):
+    inputs = tokenizer_mms(text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+    with torch.no_grad():
+        outputs = model_mms(input_ids)
+    return speech = outputs["waveform"]
+    
 def speech_to_speech_translation(audio):
     sampling_rate = 16000
     data_array,samplerate = librosa.load(audio)