app.py

import streamlit as st
import soundfile as sf
import os, re
import torch
from datautils import *
from model import Generator as Glow_model
from Hmodel import Generator as GAN_model

st.set_page_config(
    page_title = "소신 Team Demo",
    page_icon = "🔉",
)

class TTS:
    def __init__(self, model_variant):
        global device
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        torch.cuda.manual_seed(1234) if torch.cuda.is_available() else None
        self.flowgenerator = Glow_model(n_vocab = 70, h_c= 192, f_c = 768, f_c_dp = 256, out_c = 80, k_s = 3, k_s_dec = 5, heads=2, layers_enc = 6).to(device)
        self.voicegenerator = GAN_model().to(device)
        if model_variant == '감기걸린 은식':
            name = '1038_eunsik_01'
            last_chpt1 = './log/1038_eunsik_01/Glow_TTS_00289602.pt'
        elif model_variant == 'KSS':
            last_chpt1 = './log/KSS/Glow_TTS_00280641.pt'
        check_point = torch.load(last_chpt1, map_location = device)
        self.flowgenerator.load_state_dict(check_point['generator'])
        self.flowgenerator.decoder.skip()
        self.flowgenerator.eval()
        if model_variant == '감기걸린 은식':
            last_chpt2 = './log/1038_eunsik_01/HiFI_GAN_00257000.pt'
        elif model_variant == 'KSS':
            last_chpt2 = './log/KSS/HiFi_GAN_00135000.pt'
        check_point = torch.load(last_chpt2, map_location = device)
        self.voicegenerator.load_state_dict(check_point['gen_model'])
        self.voicegenerator.eval()
        self.voicegenerator.remove_weight_norm()
    
    def inference(self, input_text, noise_scale = 0.667, length_scale = 1.0):
        filters = '([.,!?])'
        sentence = re.sub(re.compile(filters), '', input_text)
        x = text_to_sequence(sentence)
        x = torch.autograd.Variable(torch.tensor(x).unsqueeze(0)).to(device).long()
        x_length = torch.tensor(x.shape[1]).unsqueeze(0).to(device)
        
        with torch.no_grad():
            (y_gen_tst, *_), *_, (attn_gen, *_) = self.flowgenerator(x, x_length, gen = True, noise_scale = noise_scale, length_scale = length_scale)
            y = self.voicegenerator(y_gen_tst)
            audio = y.squeeze() * 32768.0
            voice = audio.cpu().numpy().astype('int16')
        return voice

def init_session_state():
    # Model
    if "init_model" not in st.session_state:
        st.session_state.init_model = True
        st.session_state.model_variant = "감기걸린 은식"
        st.session_state.TTS = TTS("감기걸린 은식")

def update_model():
    if st.session_state.model_variant == "KSS":
        st.session_state.TTS = TTS("KSS")
    elif st.session_state.model_variant == "감기걸린 은식":
        st.session_state.TTS = TTS("감기걸린 은식")

def update_session_state(state_id, state_value):
    st.session_state[f"{state_id}"] = state_value
    
def centered_text(input_text, mode = "h1",):
    st.markdown(
        f"<{mode} style='text-align: center;'>{input_text}</{mode}>", unsafe_allow_html = True)

init_session_state()

centered_text("🔉 소신 Team Demo")
centered_text("mel generator : Glow-TTS, vocoder : HiFi-GAN", "h5")
st.write(" ")

mode = "p"
st.markdown(
    f"<{mode} style='text-align: left;'><small>This is a demo trained by our vocie. The voice \"KSS\" is traind by <a href= 'https://www.kaggle.com/datasets/bryanpark/korean-single-speaker-speech-dataset'>KSS Dataset</a>. The voice \"감기걸린 은식\" is trained from pre-trained \"KSS\". We got this deomoformat from Nix-TTS Interactive Demo</small></{mode}>",
    unsafe_allow_html = True
)

st.write(" ")
st.write(" ")
col1, col2 = st.columns(2)

with col1:
    input_text = st.text_input(
        "한글로만 입력해주세요",
        value = "밥은 먹고 다녀",
    )
with col2:
    model_variant = st.selectbox("목소리 선택해주세요", options = ["KSS", "감기걸린 은식"], index = 1)
    
button_change = st.button("Change Vocie") 
if button_change == True:
    if model_variant != st.session_state.model_variant:
    # Update variant choice
    update_session_state("model_variant", model_variant)
    # Re-load model
    update_model()
    st.snow()
    
noise_scale = st.slider('noise를 추가합니다.', 0., 2., value = 0.33, step = 0.01)
length_scale = st.slider('속도를 조절합니다.', 0., 2., value = 1., step = 0.01)

button_gen = st.button("Generate Voice")
if button_gen == True:
    voice = st.session_state.TTS.inference(input_text, noise_scale, length_scale)
    st.audio(voice,sample_rate = 22050)
    st.caption("Generated Voice by" + st.session_state.model_variant)
    st.balloons()