Update app.py

app.py

@@ -3,8 +3,8 @@ 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
+from transformers import pipeline
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
@@ -12,13 +12,19 @@ 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("sanchit-gandhi/speecht5_tts_vox_nl")
+# processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
+
+# model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
+# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+
+# embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+# speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
-model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+from transformers import VitsModel, VitsTokenizer
+
+model = VitsModel.from_pretrained("Matthijs/mms-tts-nld")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-nld")
 
 
 def translate(audio):
@@ -26,9 +32,17 @@ def translate(audio):
     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)
+#     return speech.cpu()
+
 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=text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+    with torch.no_grad():
+        outputs = model(input_ids)
+    speech = outputs.audio[0]
     return speech.cpu()