Spaces:
Running
on
T4
Running
on
T4
File size: 7,890 Bytes
010a382 1cc3889 010a382 1472011 010a382 925e435 010a382 d1ea58a 010a382 1472011 ef4b7af 010a382 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 |
import json
import os
import re
import librosa
import numpy as np
import torch
from torch import no_grad, LongTensor
import commons
import utils
import gradio as gr
from models import SynthesizerTrn
from text import text_to_sequence, _clean_text
from mel_processing import spectrogram_torch
limitation = os.getenv("SYSTEM") == "spaces" # limit text and audio length in huggingface spaces
def get_text(text, hps, is_phoneme):
text_norm = text_to_sequence(text, hps.symbols, [] if is_phoneme else hps.data.text_cleaners)
if hps.data.add_blank:
text_norm = commons.intersperse(text_norm, 0)
text_norm = LongTensor(text_norm)
return text_norm
def create_tts_fn(model, hps, speaker_ids):
def tts_fn(text, speaker, speed, is_phoneme):
if limitation:
text_len = len(text)
max_len = 300
if is_phoneme:
max_len *= 3
else:
if len(hps.data.text_cleaners) > 0 and hps.data.text_cleaners[0] == "zh_ja_mixture_cleaners":
text_len = len(re.sub("(\[ZH\]|\[JA\])", "", text))
if text_len > max_len:
return "Error: Text is too long", None
speaker_id = speaker_ids[speaker]
stn_tst = get_text(text, hps, is_phoneme)
with no_grad():
x_tst = stn_tst.unsqueeze(0)
x_tst_lengths = LongTensor([stn_tst.size(0)])
sid = LongTensor([speaker_id])
audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=.667, noise_scale_w=0.8,
length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
del stn_tst, x_tst, x_tst_lengths, sid
return "Success", (hps.data.sampling_rate, audio)
return tts_fn
def create_to_phoneme_fn(hps):
def to_phoneme_fn(text):
return _clean_text(text, hps.data.text_cleaners) if text != "" else ""
return to_phoneme_fn
css = """
#advanced-btn {
color: white;
border-color: black;
background: black;
font-size: .7rem !important;
line-height: 19px;
margin-top: 24px;
margin-bottom: 12px;
padding: 2px 8px;
border-radius: 14px !important;
}
#advanced-options {
display: none;
margin-bottom: 20px;
}
"""
if __name__ == '__main__':
models_tts = []
models_vc = []
models_soft_vc = []
# {"title": "γγγγ·γ―γͺγ¨γ€γγ£γ", "lang": "ζ₯ζ¬θͺ (Japanese)", "example": "γγγ«γ‘γ―γ", "type": "vits"}
name = 'γγγ»γ« TTS'
lang = 'ζ₯ζ¬θͺ (Japanese)'
example = 'γγγ«γ‘γ―γ'
config_path = f"saved_model/config.json"
model_path = f"saved_model/model.pth"
cover_path = f"saved_model/cover.png"
hps = utils.get_hparams_from_file(config_path)
model = SynthesizerTrn(
len(hps.symbols),
hps.data.filter_length // 2 + 1,
hps.train.segment_size // hps.data.hop_length,
n_speakers=hps.data.n_speakers,
**hps.model)
utils.load_checkpoint(model_path, model, None)
model.eval()
speaker_ids = [sid for sid, name in enumerate(hps.speakers) if name != "None"]
speakers = [name for sid, name in enumerate(hps.speakers) if name != "None"]
t = 'vits'
models_tts.append((name, cover_path, speakers, lang, example,
hps.symbols, create_tts_fn(model, hps, speaker_ids),
create_to_phoneme_fn(hps)))
app = gr.Blocks(css=css)
with app:
gr.Markdown("# Project Sekai TTS Using VITS Model\n\n"
"\n\n")
with gr.Tabs():
with gr.TabItem("TTS"):
with gr.Tabs():
for i, (name, cover_path, speakers, lang, example, symbols, tts_fn,
to_phoneme_fn) in enumerate(models_tts):
with gr.TabItem(f"Proseka"):
with gr.Column():
gr.Markdown(f"## {name}\n\n"
f"\n\n"
f"lang: {lang}")
tts_input1 = gr.TextArea(label="Text (150 words limitation)", value=example,
elem_id=f"tts-input{i}")
tts_input2 = gr.Dropdown(label="Speaker", choices=speakers,
type="index", value=speakers[0])
tts_input3 = gr.Slider(label="Speed", value=1, minimum=0.1, maximum=2, step=0.1)
with gr.Accordion(label="Advanced Options", open=False):
phoneme_input = gr.Checkbox(value=False, label="Phoneme input")
to_phoneme_btn = gr.Button("Covert text to phoneme")
phoneme_list = gr.Dataset(label="Phoneme list", components=[tts_input1],
samples=[[x] for x in symbols],
elem_id=f"phoneme-list{i}")
phoneme_list_json = gr.Json(value=symbols, visible=False)
tts_submit = gr.Button("Generate", variant="primary")
tts_output1 = gr.Textbox(label="Output Message")
tts_output2 = gr.Audio(label="Output Audio")
tts_submit.click(tts_fn, [tts_input1, tts_input2, tts_input3, phoneme_input],
[tts_output1, tts_output2])
to_phoneme_btn.click(to_phoneme_fn, [tts_input1], [tts_input1])
phoneme_list.click(None, [phoneme_list, phoneme_list_json], [],
_js=f"""
(i,phonemes) => {{
let root = document.querySelector("body > gradio-app");
if (root.shadowRoot != null)
root = root.shadowRoot;
let text_input = root.querySelector("#tts-input{i}").querySelector("textarea");
let startPos = text_input.selectionStart;
let endPos = text_input.selectionEnd;
let oldTxt = text_input.value;
let result = oldTxt.substring(0, startPos) + phonemes[i] + oldTxt.substring(endPos);
text_input.value = result;
let x = window.scrollX, y = window.scrollY;
text_input.focus();
text_input.selectionStart = startPos + phonemes[i].length;
text_input.selectionEnd = startPos + phonemes[i].length;
text_input.blur();
window.scrollTo(x, y);
return [];
}}""")
gr.Markdown(
"Official User Page \n\n"
"- [https://github.com/kdrkdrkdr/ProsekaTTS](https://github.com/kdrkdrkdr/ProsekaTTS)\n\n"
"Reference \n\n"
"- [https://huggingface.co/spaces/skytnt/moe-tts](https://huggingface.co/spaces/skytnt/moe-tts)"
)
app.queue(concurrency_count=3).launch(show_api=False)
|