demo

app.py

@@ -4,12 +4,9 @@ import librosa
 import numpy as np
 import torch
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5Processor, SpeechT5HifiGan
+from transformers import pipeline
 
-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")
+synthesiser = pipeline("text-to-speech", "techiaith/microsoft_speecht5_finetuned_bu_tts_cy_en")
 
 speaker_embeddings = {
     "GGP": "spkemb/speaker0.npy",
@@ -41,8 +38,7 @@ def predict(text, speaker):
     speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
     speaker_embedding = prepare_dataset(speaker_embedding)
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
+    speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
     speech = (speech.numpy() * 32767).astype(np.int16)
     return (16000, speech)