Trying WasuratS' code

app.py

@@ -12,25 +12,22 @@ 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")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
+processor = SpeechT5Processor.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl")
+model = SpeechT5ForTextToSpeech.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_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)
 
-
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
-    return outputs["text"]
-
-
+    return outputs['text']
+    
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")
-    # Reducing the input length to 600
-    inputs = inputs["input_ids"][:600]
-    speech = model.generate_speech(inputs.to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    input_ids = inputs["input_ids"]
+    input_ids = input_ids[0:1, :200]
+    speech = model.generate_speech(input_ids.to(device), speaker_embeddings.to(device), vocoder=vocoder)
     return speech.cpu()