Spaces:
Sleeping
Sleeping
| from spleeter.separator import Separator | |
| import gradio as gr | |
| import shutil | |
| import librosa | |
| import soundfile | |
| import os | |
| from gradio_client import Client, file | |
| import subprocess | |
| try: | |
| shutil.rmtree("output") | |
| except FileNotFoundError: | |
| pass | |
| def spleeter(aud, model_file, index_file, pitch): | |
| separator = Separator('spleeter:2stems') | |
| filename = os.path.basename(aud).split('.')[0] | |
| accompaniment_filename = f"./output/audio_example/{filename}_accompaniment.wav" | |
| vocal_filename = f"./output/audio_example/{filename}_vocals.wav" | |
| song_filename = f'./output/audio_example/{filename}_song.wav' | |
| separator.separate_to_file(aud, "output/", filename_format="audio_example/%s_{instrument}.wav" % filename) | |
| # change pitch | |
| y, sr = librosa.load(accompaniment_filename) | |
| new_accompaniment = librosa.effects.pitch_shift(y, sr, pitch) | |
| soundfile.write(accompaniment_filename, new_accompaniment, sr, ) | |
| # send vocal for processing | |
| # https://huggingface.co/spaces/r3gm/rvc_zero | |
| client = Client("r3gm/rvc_zero") | |
| result = client.predict( | |
| audio_files=[file(vocal_filename)], | |
| file_m=file(model_file), | |
| pitch_alg="rmvpe+", | |
| pitch_lvl=pitch, | |
| file_index=file(index_file), | |
| index_inf=0.75, | |
| r_m_f=3, | |
| e_r=0.25, | |
| c_b_p=0.5, | |
| active_noise_reduce=False, | |
| audio_effects=False, | |
| api_name="/run" | |
| )[0] | |
| # combine accompaniment and vocal | |
| command = f"""ffmpeg -i "{result}" -i "{accompaniment_filename}" -filter_complex amix=inputs=2:duration=longest "{song_filename}" """ | |
| subprocess.call(command, shell=True) | |
| return song_filename | |
| inputs = [ | |
| gr.Audio(sources='upload', label="Input Audio File", type="filepath"), | |
| gr.File( | |
| label="Model file", | |
| type="filepath", | |
| height=130, | |
| ), | |
| gr.File( | |
| label="Index file", | |
| type="filepath", | |
| height=130, | |
| ), | |
| gr.Slider( | |
| label="Pitch level", | |
| minimum=-24, | |
| maximum=24, | |
| step=1, | |
| value=0, | |
| visible=True, | |
| interactive=True, | |
| ) | |
| ] | |
| outputs = [ | |
| gr.Audio(label="Output Audio", type="filepath"), | |
| ] | |
| title = "Music Spleeter" | |
| description = "Clearing a musical composition of the performer's voice is a common task. It is solved well, for example, by professional audio file editing programs. AI algorithms have also been gaining ground recently." | |
| article = "<div style='text-align: center; max-width:800px; margin:10px auto;'><p>In this case we use Deezer's Spleeter with ready pretrained models. It can leave as an output both just the music and just the performer's voice.</p><p>Sources: <a href='https://github.com/deezer/spleeter/' target='_blank'>Spleeter</a>: a Fast and Efficient Music Source Separation Tool with Pre-Trained Models</p><p style='text-align: center'><a href='https://starstat.yt/cat/music' target='_blank'>StarStat Music</a>: Youtubers Net Worth in category Music</p></div>" | |
| if __name__ == '__main__': | |
| gr.Interface(spleeter, inputs, outputs).launch() | |