Update app.py

app.py

@@ -6,7 +6,7 @@ from datasets import load_dataset
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
 
 ## Imports for MMS
-# from transformers import VitsModel, VitsTokenizer
+from transformers import VitsModel, VitsTokenizer
 
 
 
@@ -25,9 +25,16 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
 # For Dutch
 
 ##### speecht5 #####
-model_id = 'sanchit-gandhi/speecht5_tts_vox_nl'
-processor = SpeechT5Processor.from_pretrained(model_id)
-model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
+# model_id = 'sanchit-gandhi/speecht5_tts_vox_nl'
+# processor = SpeechT5Processor.from_pretrained(model_id)
+# model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
+
+# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+
+
+##### mms #####
+model = VitsModel.from_pretrained("Matthijs/mms-tts-nld")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-nld")
 
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
@@ -35,6 +42,8 @@ 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)
 
@@ -48,11 +57,21 @@ def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]
 
+# Original
+# 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, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model(inputs["input_ids"])
+    speech = outputs.audio[0]
+    
     return speech.cpu()
+    
 
 
 def speech_to_speech_translation(audio):