Synthesize nl

app.py

@@ -4,6 +4,8 @@ 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"
@@ -14,7 +16,10 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+# model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = VitsModel.from_pretrained("Matthijs/mms-tts-nl")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-nl")
+
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
@@ -25,9 +30,10 @@ 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")
+    inputs = tokenizer(text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+    # 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()