app.py

import streamlit as st
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import json
import os
import tempfile
import shutil
import requests
from pathlib import Path
temp_dir = tempfile.TemporaryDirectory()
global ckpt_temp_file
global audio_temp_file
global config_temp_file
###################################################
from utils.hparams import hparams
from preprocessing.data_gen_utils import get_pitch_parselmouth,get_pitch_crepe
import numpy as np
import matplotlib.pyplot as plt
import IPython.display as ipd
import utils
import librosa
import torchcrepe
from infer import *
import logging
from infer_tools.infer_tool import *
import io
import parselmouth

clip_completed = False
def render_audio(ckpt_temp_file, config_temp_file, audio_temp_file, title, title2, title3, choice, noise_step, use_mel_as_base):
    logging.getLogger('numba').setLevel(logging.WARNING)
    title = int(title)
    title2 = int(title2)
    project_name = "Unnamed"
    model_path = ckpt_temp_file
    config_path= config_temp_file
    hubert_gpu=True
    svc_model = Svc(project_name,config_path,hubert_gpu, model_path)
    print('model loaded')
    wav_fn = audio_temp_file
    demoaudio, sr = librosa.load(wav_fn)
    key = title # 音高调整，支持正负（半音）
    # 加速倍数
    pndm_speedup = 20
    wav_gen='que.wav'
    
    # Show the spinner and run the run_clip function inside the 'with' block
    with st.spinner("Rendering Audio..."):
      f0_tst, f0_pred, audio = run_clip(svc_model,file_path=wav_fn, key=key, acc=title2, use_crepe=choice, use_pe=False, thre=0.05,
                                        use_gt_mel=use_mel_as_base, add_noise_step=noise_step,project_name=project_name,out_path=wav_gen)
##PRAAT
    formant_shift_ratio_str = title3
    formant_shift_ratio = float(formant_shift_ratio_str)
    
    # If the formant shift ratio is not equal to 1.0, change the gender of the sound using parselmouth
    if formant_shift_ratio != 1.0:
        sound = parselmouth.Sound(wav_gen)
        print(wav_gen)
        Audio(data=sound.values, rate=sound.sampling_frequency)
        sound.get_power()
        sampling_rate = sound.sampling_frequency
        print(sampling_rate)
        resampled_sound = sound.resample(sampling_rate)
        print(resampled_sound)
        factor = formant_shift_ratio
        print(factor)
        manipulated_sound = call(sound, "Change gender", 75, 500, factor, 0, 1, 1)
        print(manipulated_sound)
        manipulated_sound.save("que.wav", "WAV")
        print("Gender correct!")
    clip_completed = True
    if clip_completed:
        st.audio(wav_gen)

#######################################################
st.set_page_config(
    page_title="DiffSVC Render",
    page_icon="🧊",
    initial_sidebar_state="expanded",
)
############
col1, col2 = st.columns(2)
col1.title('DIFF-SVC Render')
col2.title('Settings')
ckpt = col1.file_uploader("Choose your CKPT", type='ckpt')
if ckpt is not None:
  #TEMP FUNCTION
  with tempfile.NamedTemporaryFile(mode="wb", suffix='.ckpt', delete=False) as temp:
    # Get the file contents as bytes
    bytes_data = ckpt.getvalue()
    # Write the bytes to the temporary file
    temp.write(bytes_data)
    ckpt_temp_file = temp.name
    # Print the temporary file name
    print(temp.name)
# Display the file path
if "ckpt_temp_file" in locals():
    st.success("File saved to: {}".format(ckpt_temp_file))
# File uploader
config = col1.file_uploader("Choose your config", type='yaml')
if config is not None:
  #TEMP FUNCTION
  with tempfile.NamedTemporaryFile(mode="wb", suffix='.yaml', delete=False) as temp:
    # Get the file contents as bytes
    bytes_data = config.getvalue()
    # Write the bytes to the temporary file
    temp.write(bytes_data)
    config_temp_file = temp.name
    # Print the temporary file name
    print(temp.name)
# Display the file path
if "config_temp_file" in locals():
    st.success("File saved to: {}".format(config_temp_file))

audio = col1.file_uploader("Choose your audio", type=["wav"])
if audio is not None:
#EMP FUNCTION
  with tempfile.NamedTemporaryFile(mode="wb", suffix='.wav', delete=False) as temp:
# Get the file contents as bytes
    bytes_data = audio.getvalue()
# Write the bytes to the temporary file
    temp.write(bytes_data)
    audio_temp_file = temp.name
# Print the temporary file name
    print(temp.name)
# Display the file path
if "audio_temp_file" in locals():
    st.success("File saved to: {}".format(audio_temp_file))

title = col2.number_input("Key", value=0, step=1, min_value=-12, max_value=12)
title2 = col2.number_input("Speedup", value=20, step=1, min_value=5, max_value=100)
title3 = col2.number_input("Gender Flag", value=1.00, step=0.01, min_value=0.70, max_value=1.30, help='Default is 1.0, it works by decimals, setting it at 1.05 will make your render sound more female-ish, setting it to 0.95 will make it sound more masculine, for example.')
choice = col2.checkbox('Use Crepe', value=False)
# Create checkbox for using Mel as Base
use_mel_as_base = col2.checkbox('Use Mel as Base', value=False, help='gt mel: Enabling this will use the input audio as a base and will unlock a new parameter, do not use this if you dont know what it does.')
noise_step = 600
# Show "Noise Step" input parameter when checkbox is checked
if use_mel_as_base:
    noise_step = col2.number_input('Noise Step', value=600, min_value=1, max_value=1000, step=50)
else:
    noise_step = 600
password = col2.text_input("Enter password", help='Password can be got by agreeing to TOS and getting allowed after validation, you can go to the TOS here:')
correct_password = os.environ.get("gatepassword")
###
if st.button("Render audio"):
    if password == correct_password:
        render_audio(ckpt_temp_file, config_temp_file, audio_temp_file, title, title2, title3, noise_step, choice, use_mel_as_base)
  
    else:
        st.error("Incorrect password")