import gradio as gr import torch from datasets import load_dataset from transformers import pipeline, SpeechT5Processor, SpeechT5HifiGan, SpeechT5ForTextToSpeech model_id = "Sandiago21/speecht5_finetuned_voxpopuli_it" # update with your model id # pipe = pipeline("automatic-speech-recognition", model=model_id) model = SpeechT5ForTextToSpeech.from_pretrained(model_id) vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan") embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation") speaker_embeddings = torch.tensor(embeddings_dataset[7440]["xvector"]).unsqueeze(0) checkpoint = "microsoft/speecht5_tts" processor = SpeechT5Processor.from_pretrained(checkpoint) replacements = [ ("á", "a"), ("ç", "c"), ("è", "e"), ("ì", "i"), ("í", "i"), ("ò", "o"), ("ó", "o"), ("ù", "u"), ("ú", "u"), ("š", "s"), ("ï", "i"), ] title = "Text-to-Speech" description = """ Demo for text-to-speech translation in Italian. Demo uses [Sandiago21/speecht5_finetuned_voxpopuli_it](https://huggingface.co/Sandiago21/speecht5_finetuned_voxpopuli_it) checkpoint, which is based on Microsoft's [SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model and is fine-tuned in Italian Audio dataset ![Text-to-Speech (TTS)"](https://geekflare.com/wp-content/uploads/2021/07/texttospeech-1200x385.png "Diagram of Text-to-Speech (TTS)") """ def cleanup_text(text): for src, dst in replacements: text = text.replace(src, dst) return text def synthesize_speech(text): text = cleanup_text(text) inputs = processor(text=text, return_tensors="pt") speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder) return gr.Audio.update(value=(16000, speech.cpu().numpy())) syntesize_speech_gradio = gr.Interface( synthesize_speech, inputs = gr.Textbox(label="Text", placeholder="Type something here..."), outputs=gr.Audio(), examples=["Grandi manifestazioni in polonia al momento della firma dell'accordo che hanno portato il governo polacco a decidere di non ratificare l'accordo per il momento"], title=title, description=description, ).launch()