Spaces:
Running
Running
File size: 3,493 Bytes
e976963 a1ea115 e976963 a1ea115 e976963 a1ea115 e976963 3b09304 e976963 a1ea115 e976963 a1ea115 e976963 a1ea115 e976963 21b151b e976963 a1ea115 e976963 a1ea115 e976963 a1ea115 e976963 3b09304 e976963 3b09304 e976963 3b09304 e976963 a1ea115 e976963 3b09304 e976963 a1ea115 |
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 |
import gc
import hashlib
import os
import shlex
import subprocess
import librosa
import torch
import numpy as np
import soundfile as sf
import gradio as gr
from rvc import Config, load_hubert, get_vc, rvc_infer
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
RVC_MODELS_DIR = os.path.join(BASE_DIR, 'rvc_models')
OUTPUT_DIR = os.path.join(BASE_DIR, 'song_output')
def get_rvc_model(voice_model):
model_dir = os.path.join(RVC_MODELS_DIR, voice_model)
rvc_model_path = next((os.path.join(model_dir, f) for f in os.listdir(model_dir) if f.endswith('.pth')), None)
rvc_index_path = next((os.path.join(model_dir, f) for f in os.listdir(model_dir) if f.endswith('.index')), None)
if rvc_model_path is None:
raise FileNotFoundError(f'There is no model file in the {model_dir} directory.')
return rvc_model_path, rvc_index_path
def convert_to_stereo(audio_path):
wave, sr = librosa.load(audio_path, mono=False, sr=44100)
if type(wave[0]) != np.ndarray:
stereo_path = 'Voice_stereo.wav'
command = shlex.split(f'ffmpeg -y -loglevel error -i "{audio_path}" -ac 2 -f wav "{stereo_path}"')
subprocess.run(command)
return stereo_path
return audio_path
def get_hash(filepath):
file_hash = hashlib.blake2b()
with open(filepath, 'rb') as f:
while chunk := f.read(8192):
file_hash.update(chunk)
return file_hash.hexdigest()[:11]
def display_progress(percent, message, progress=gr.Progress()):
progress(percent, desc=message)
def voice_change(voice_model, vocals_path, output_path, pitch_change, f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length):
rvc_model_path, rvc_index_path = get_rvc_model(voice_model)
device = 'cpu'
config = Config(device, True)
hubert_model = load_hubert(device, config.is_half, os.path.join(RVC_MODELS_DIR, 'hubert_base.pt'))
cpt, version, net_g, tgt_sr, vc = get_vc(device, config.is_half, config, rvc_model_path)
rvc_infer(rvc_index_path, index_rate, vocals_path, output_path, pitch_change, f0_method, cpt, version, net_g,
filter_radius, tgt_sr, rms_mix_rate, protect, crepe_hop_length, vc, hubert_model)
del hubert_model, cpt, net_g, vc
gc.collect()
torch.cuda.empty_cache()
def song_cover_pipeline(uploaded_file, voice_model, pitch_change, index_rate=0.5, filter_radius=3, rms_mix_rate=0.25, f0_method='rmvpe',
crepe_hop_length=128, protect=0.33, output_format='mp3', progress=gr.Progress()):
if not uploaded_file or not voice_model:
raise ValueError('Make sure that the song input field and voice model field are filled in.')
display_progress(0, '[~] Starting the AI cover generation pipeline...', progress)
if not os.path.exists(uploaded_file):
raise FileNotFoundError(f'{uploaded_file} does not exist.')
song_id = get_hash(uploaded_file)
song_dir = os.path.join(OUTPUT_DIR, song_id)
os.makedirs(song_dir, exist_ok=True)
orig_song_path = convert_to_stereo(uploaded_file)
ai_cover_path = os.path.join(song_dir, f'Converted_Voice.{output_format}')
if os.path.exists(ai_cover_path):
os.remove(ai_cover_path)
display_progress(0.5, '[~] Converting vocals...', progress)
voice_change(voice_model, orig_song_path, ai_cover_path, pitch_change, f0_method, index_rate,
filter_radius, rms_mix_rate, protect, crepe_hop_length)
return ai_cover_path |