# coding=utf-8 import time import os import gradio as gr import utils import argparse import commons from models import SynthesizerTrn from text import text_to_sequence import torch from torch import no_grad, LongTensor import webbrowser import logging import gradio.processing_utils as gr_processing_utils logging.getLogger('numba').setLevel(logging.WARNING) limitation = os.getenv("SYSTEM") == "spaces" # limit text and audio length in huggingface spaces audio_postprocess_ori = gr.Audio.postprocess def audio_postprocess(self, y): data = audio_postprocess_ori(self, y) if data is None: return None return gr_processing_utils.encode_url_or_file_to_base64(data["name"]) gr.Audio.postprocess = audio_postprocess def get_text(text, hps): text_norm, clean_text = text_to_sequence(text, hps.symbols, hps.data.text_cleaners) if hps.data.add_blank: text_norm = commons.intersperse(text_norm, 0) text_norm = LongTensor(text_norm) return text_norm, clean_text def vits(text, language, speaker_id, noise_scale, noise_scale_w, length_scale): start = time.perf_counter() if not len(text): return "输入文本不能为空!", None, None text = text.replace('\n', ' ').replace('\r', '').replace(" ", "") if len(text) > 100 and limitation: return f"输入文字过长!{len(text)}>100", None, None if language == 0: text = f"[ZH]{text}[ZH]" elif language == 1: text = f"[JA]{text}[JA]" else: text = f"{text}" stn_tst, clean_text = get_text(text, hps_ms) with no_grad(): x_tst = stn_tst.unsqueeze(0).to(device) x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device) speaker_id = LongTensor([speaker_id]).to(device) audio = net_g_ms.infer(x_tst, x_tst_lengths, sid=speaker_id, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale)[0][0, 0].data.cpu().float().numpy() return "生成成功!", (22050, audio), f"生成耗时 {round(time.perf_counter()-start, 2)} s" def search_speaker(search_value): for s in speakers: if search_value == s: return s for s in speakers: if search_value in s: return s def change_lang(language): if language == 0: return 0.6, 0.668, 1.2 else: return 0.6, 0.668, 1.1 download_audio_js = """ () =>{{ let root = document.querySelector("body > gradio-app"); if (root.shadowRoot != null) root = root.shadowRoot; let audio = root.querySelector("#tts-audio").querySelector("audio"); let text = root.querySelector("#input-text").querySelector("textarea"); if (audio == undefined) return; text = text.value; if (text == undefined) text = Math.floor(Math.random()*100000000); audio = audio.src; let oA = document.createElement("a"); oA.download = text.substr(0, 20)+'.wav'; oA.href = audio; document.body.appendChild(oA); oA.click(); oA.remove(); }} """ if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--device', type=str, default='cpu') parser.add_argument('--api', action="store_true", default=False) parser.add_argument("--share", action="store_true", default=False, help="share gradio app") parser.add_argument("--colab", action="store_true", default=False, help="share gradio app") args = parser.parse_args() device = torch.device(args.device) hps_ms = utils.get_hparams_from_file(r'./model/config.json') net_g_ms = SynthesizerTrn( len(hps_ms.symbols), hps_ms.data.filter_length // 2 + 1, hps_ms.train.segment_size // hps_ms.data.hop_length, n_speakers=hps_ms.data.n_speakers, **hps_ms.model) _ = net_g_ms.eval().to(device) speakers = hps_ms.speakers model, optimizer, learning_rate, epochs = utils.load_checkpoint(r'./model/G_latest.pth', net_g_ms, None) with gr.Blocks() as app: gr.Markdown( "#