Update app.py

app.py

@@ -2,35 +2,26 @@ import gradio as gr
 import numpy as np
 import torch
 from datasets import load_dataset
-
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from espnet2.bin.tts_inference import Text2Speech
+from transformers import pipeline
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").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)
-
+tts_model = Text2Speech.from_pretrained("espnet/kan-bayashi_ljspeech_tts_train_tacotron2_raw_phn_tacotron_g2p_en_no_space_train.loss.best")
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["text"]
 
-
 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()
-
+    with torch.no_grad():
+        wav = tts_model(text)["wav"]
+    return wav.view(-1).cpu()
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)