demo

app.py

@@ -9,6 +9,9 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5Processor, SpeechT5Hif
 checkpoint = "microsoft/speecht5_tts"
 processor = SpeechT5Processor.from_pretrained(checkpoint)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
+model = SpeechT5ForTextToSpeech.from_pretrained(
+    "techiaith/microsoft_speecht5_finetuned_bu_tts_cy_en"
+)
 
 speaker_embeddings = {
     "GGP": "spkemb/speaker0.npy",
@@ -23,9 +26,6 @@ def predict(text, speaker):
 
     speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
-    model = SpeechT5ForTextToSpeech.from_pretrained(
-        "techiaith/microsoft_speecht5_finetuned_bu_tts_cy_en"
-    )
     inputs = processor(text=text, return_tensors="pt")
     speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
     speech = (speech.numpy() * 32767).astype(np.int16)