changed to run speecht5

app.py

@@ -3,19 +3,25 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import VitsModel, VitsTokenizer, pipeline
-
+#from transformers import VitsModel, VitsTokenizer, pipeline
+from transformers import pipeline, SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
+embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+
 # load speech translation checkpoint
 pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint for MMS
-model = VitsModel.from_pretrained("Matthijs/mms-tts-fra")
-tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-fra")
+# model = VitsModel.from_pretrained("Matthijs/mms-tts-fra")
+# tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-fra")
 
+processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
+model = SpeechT5ForTextToSpeech.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_french")
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
 
 def translate(audio):
     outputs = pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "fr"})
@@ -23,9 +29,9 @@ def translate(audio):
 
 
 def synthesise(text):
-    inputs = tokenizer(text=text, return_tensors="pt")
-    speech = model(inputs["input_ids"].to(device)) 
-    return speech.audio[0].cpu()
+    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 speech_to_speech_translation(audio):