Spaces:
Runtime error
Runtime error
File size: 3,537 Bytes
11120b4 9070b28 11120b4 9070b28 11120b4 b17c40c 11120b4 48157e5 11120b4 66de3a6 11120b4 b3d59f4 11120b4 66de3a6 |
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 |
import os
import sys
import json
from subprocess import call
import torch
import gradio as gr
from scipy.io.wavfile import write
from huggingface_hub import hf_hub_url, cached_download
import nltk
from nltk.tokenize import word_tokenize
nltk.download("punkt")
AUTH_TOKEN = os.environ["HF_TOKEN"]
# download models
url = hf_hub_url(repo_id="bookbot/grad-tts-en-ft-Weildan-v2", filename="grad_24000.pt")
grad_tts_model_path = cached_download(url, use_auth_token=AUTH_TOKEN)
torch.hub.download_url_to_file(
"https://github.com/AK391/Speech-Backbones/releases/download/v1/hifigan.pt",
"hifigan.pt",
)
# build MAS
current = os.getcwd()
os.chdir(current + "/Grad-TTS/model/monotonic_align")
call("python setup.py build_ext --inplace", shell=True)
os.chdir("../../../")
sys.path.append("Grad-TTS/")
import params
from model import GradTTS
from text import text_to_sequence, cmudict
from text.symbols import symbols
from utils import intersperse
sys.path.append("Grad-TTS/hifi-gan/")
from env import AttrDict
from models import Generator as HiFiGAN
SPEAKERS = 247
# load models
generator = GradTTS(
len(symbols) + 1,
SPEAKERS,
params.spk_emb_dim,
params.n_enc_channels,
params.filter_channels,
params.filter_channels_dp,
params.n_heads,
params.n_enc_layers,
params.enc_kernel,
params.enc_dropout,
params.window_size,
params.n_feats,
params.dec_dim,
params.beta_min,
params.beta_max,
pe_scale=1000,
)
generator.load_state_dict(
torch.load(grad_tts_model_path, map_location=lambda loc, storage: loc)
)
_ = generator.eval()
cmu = cmudict.CMUDict("./Grad-TTS/resources/cmu_dictionary_id")
with open("./Grad-TTS/checkpts/hifigan-config.json") as f:
h = AttrDict(json.load(f))
hifigan = HiFiGAN(h)
hifigan.load_state_dict(
torch.load("./hifigan.pt", map_location=lambda loc, storage: loc)["generator"]
)
_ = hifigan.eval()
hifigan.remove_weight_norm()
def inference(text, n_timesteps):
text = " ".join(word_tokenize(text))
x = torch.LongTensor(
intersperse(text_to_sequence(text, dictionary=cmu), len(symbols))
)[None]
x_lengths = torch.LongTensor([x.shape[-1]])
y_enc, y_dec, attn = generator.forward(
x,
x_lengths,
n_timesteps=n_timesteps,
temperature=1.5,
stoc=False,
spk=torch.LongTensor([0]) if SPEAKERS > 1 else None,
length_scale=1.0,
)
with torch.no_grad():
audio = hifigan.forward(y_dec).cpu().squeeze().clamp(-1, 1).detach().numpy()
write("out.wav", 22050, audio)
return "./out.wav"
inputs = [
gr.inputs.Textbox(lines=5, label="Input Text"),
gr.inputs.Slider(minimum=0, maximum=100, step=10, label="Timesteps"),
]
outputs = gr.outputs.Audio(type="file", label="Output Audio")
title = "Bookbot Grad-TTS Weildan Demo 🐨"
description = "Hi there! You can start typing any input text ⌨, select your preferred timesteps ⌚, and hit submit! 🚂 Please be patient as it may take a while - the greater the timestep, the longer the generation 😁"
utterances = [
"Selamat pagi! Selamat datang di Jakarta!",
"Kak, harga nasi gorengnya berapa ya?",
"Bapak bilang, Malik hebat. Bisa bersih bersih seperti Bapak.",
"Here are the match lineups for the Colombia Haiti match.",
]
timesteps = [(i * 10) + 50 for i in range(len(utterances))]
examples = [list(l) for l in zip(utterances, timesteps)]
gr.Interface(
inference, inputs, outputs, title=title, description=description, examples=examples
).launch()
|