|
|
import os |
|
|
import gradio as gr |
|
|
from lib.infer import infer_audio |
|
|
|
|
|
from pydub import AudioSegment |
|
|
import zipfile |
|
|
import shutil |
|
|
import urllib.request |
|
|
import gdown |
|
|
|
|
|
main_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) |
|
|
|
|
|
os.chdir(main_dir) |
|
|
|
|
|
def upload_audio(model_name, sound_path, f0_change, f0_method, min_pitch, max_pitch, |
|
|
crepe_hop_length, index_rate, filter_radius, rms_mix_rate, |
|
|
protect, split_infer, min_silence, silence_threshold, |
|
|
seek_step, keep_silence, formant_shift, quefrency, timbre, |
|
|
f0_autotune, output_format): |
|
|
|
|
|
if not sound_path: |
|
|
uploaded_audio = files.upload() |
|
|
assert len(uploaded_audio) == 1, "Please only input audio one at a time" |
|
|
sound_path = os.path.join(os.getcwd(), list(uploaded_audio.keys())[0]) |
|
|
|
|
|
inferred_audio = infer_audio( |
|
|
model_name, |
|
|
sound_path, |
|
|
f0_change, |
|
|
f0_method, |
|
|
min_pitch, |
|
|
max_pitch, |
|
|
crepe_hop_length, |
|
|
index_rate, |
|
|
filter_radius, |
|
|
rms_mix_rate, |
|
|
protect, |
|
|
split_infer, |
|
|
min_silence, |
|
|
silence_threshold, |
|
|
seek_step, |
|
|
keep_silence, |
|
|
formant_shift, |
|
|
quefrency, |
|
|
timbre, |
|
|
f0_autotune, |
|
|
output_format |
|
|
) |
|
|
|
|
|
return AudioSegment.from_file(inferred_audio) |
|
|
|
|
|
def download_model(url, dir_name): |
|
|
models_dir = "models" |
|
|
extraction_folder = os.path.join(models_dir, dir_name) |
|
|
|
|
|
if os.path.exists(extraction_folder): |
|
|
return f'Voice model directory {dir_name} already exists! Choose a different name.' |
|
|
|
|
|
if 'pixeldrain.com' in url: |
|
|
zip_name = url.split('/')[-1] |
|
|
url = f'https://pixeldrain.com/api/file/{zip_name}' |
|
|
elif 'drive.google.com' in url: |
|
|
zip_name = dir_name + ".zip" |
|
|
gdown.download(url, output=zip_name, use_cookies=True, quiet=True, fuzzy=True) |
|
|
else: |
|
|
zip_name = url.split('/')[-1] |
|
|
urllib.request.urlretrieve(url, zip_name) |
|
|
|
|
|
with zipfile.ZipFile(zip_name, 'r') as zip_ref: |
|
|
zip_ref.extractall(extraction_folder) |
|
|
|
|
|
os.remove(zip_name) |
|
|
|
|
|
return f'{dir_name} model successfully downloaded!' |
|
|
|
|
|
with gr.Blocks() as app: |
|
|
gr.Markdown("## Inference") |
|
|
|
|
|
with gr.Row(): |
|
|
model_name = gr.Textbox(label="Model Name") |
|
|
sound_path = gr.Audio(label="Audio Path") |
|
|
|
|
|
with gr.Row(): |
|
|
f0_change = gr.Slider(minimum=-12, maximum=12, label="F0 Change (semitones)", value=0) |
|
|
f0_method = gr.Dropdown(choices=["crepe", "harvest", "mangio-crepe", "rmvpe", "rmvpe+", "fcpe", "fcpe_legacy", "hybrid[mangio-crepe+rmvpe]", "hybrid[mangio-crepe+fcpe]", "hybrid[rmvpe+fcpe]", "hybrid[mangio-crepe+rmvpe+fcpe]"], label="F0 Method", value="fcpe") |
|
|
|
|
|
|
|
|
|
|
|
output_format = gr.Dropdown(choices=["wav", "flac", "mp3"], label="Output Format", value="wav") |
|
|
|
|
|
|
|
|
output_audio = gr.Audio(label="Inferred Audio Output") |
|
|
|
|
|
|
|
|
with gr.Tab("Download Models"): |
|
|
url = gr.Textbox(label="Model Download URL") |
|
|
dir_name = gr.Textbox(label="Desired Model Name") |
|
|
download_button = gr.Button("Download Model") |
|
|
download_output = gr.Textbox(label="output") |
|
|
|
|
|
submit_button = gr.Button("Infer") |
|
|
download_button.click(download_model, inputs=[url, dir_name], outputs=download_output) |
|
|
submit_button.click(upload_audio, inputs=[model_name, sound_path, f0_change, f0_method, output_format], outputs=output_audio) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
app.launch() |
|
|
|
