Spaces:
Sleeping
Sleeping
| import datasets | |
| from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan | |
| from datasets import load_dataset | |
| import torch | |
| import soundfile as sf | |
| import numpy as np | |
| import gradio as gr | |
| import io | |
| import sentencepiece | |
| # Charger les modèles et les embeddings du locuteur une seule fois pour éviter de les recharger à chaque appel | |
| processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts") | |
| model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts") | |
| vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan") | |
| embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation") | |
| speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0) | |
| def text_to_speech(text): | |
| # Prétraiter le texte | |
| inputs = processor(text=text, return_tensors="pt") | |
| # Générer la parole | |
| speech = model.generate_speech( | |
| inputs["input_ids"], speaker_embeddings, vocoder=vocoder | |
| ) | |
| # Enregistrer l'audio dans un buffer | |
| buffer = io.BytesIO() | |
| sf.write(buffer, speech.numpy(), samplerate=16000, format="WAV") | |
| return buffer.getvalue() | |
| # Créer l'interface Gradio | |
| interface = gr.Interface( | |
| fn=text_to_speech, | |
| inputs="text", | |
| outputs=gr.Audio(label="Processed Audio"), | |
| title="Application du type Text to speech", | |
| description="Entrez un texte en anglais et l'application va la traduire en audio" | |
| ).launch() |