Spaces:
Build error
Build error
File size: 13,575 Bytes
cbe698e e316ba7 cbe698e e316ba7 cbe698e e316ba7 cbe698e e316ba7 e586364 e316ba7 e586364 cbe698e 6ab4004 cbe698e e316ba7 cbe698e 6ab4004 e586364 cbe698e e586364 cbe698e d84fb5f cbe698e 6ab4004 e316ba7 6ab4004 cbe698e e316ba7 cbe698e d9755fb cbe698e 6ab4004 cbe698e d9755fb 6ab4004 d9755fb 6ab4004 d9755fb 6ab4004 d9755fb 6ab4004 d84fb5f e316ba7 d84fb5f e316ba7 d84fb5f cbe698e e316ba7 e586364 cbe698e e316ba7 cbe698e e316ba7 e586364 e316ba7 cbe698e e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 cbe698e e316ba7 cbe698e e316ba7 e586364 e316ba7 cbe698e e316ba7 cbe698e e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 e316ba7 e586364 cbe698e e316ba7 e586364 cbe698e e586364 d9755fb e316ba7 bb1950e cbe698e d84fb5f |
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 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 |
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# Updated to account for UI changes from https://github.com/rkfg/audiocraft/blob/long/app.py
# also released under the MIT license.
import argparse
from concurrent.futures import ProcessPoolExecutor
import os
from pathlib import Path
import subprocess as sp
from tempfile import NamedTemporaryFile
import time
import typing as tp
import warnings
import glob
import torch
import gradio as gr
import numpy as np
from audiocraft.data.audio_utils import convert_audio
from audiocraft.data.audio import audio_write, audio_read
from audiocraft.models import MusicGen
from demucs import pretrained
from demucs.apply import apply_model
from demucs.audio import convert_audio
from gradio_client import Client
import pretty_midi
LOCAL = False
USE_MIDI = True
MODEL = None # Last used model
DEMUCS_MODEL = None
MAX_BATCH_SIZE = 12
INTERRUPTING = False
client = None
# We have to wrap subprocess call to clean a bit the log when using gr.make_waveform
_old_call = sp.call
stem2idx = {'drums': 0, 'bass': 1, 'other': 2, 'vocal': 3}
stem_idx = torch.LongTensor([stem2idx['vocal'], stem2idx['other'], stem2idx['bass']])
melody_files = list(glob.glob('clips/**/*.wav', recursive=True))
midi_files = list(glob.glob('clips/**/*.mid', recursive=True))
crops = [(0, 5), (0, 10), (0, 15)]
selected_melody = ""
selected_crop = None
selected_text = ""
def _call_nostderr(*args, **kwargs):
# Avoid ffmpeg vomitting on the logs.
kwargs['stderr'] = sp.DEVNULL
kwargs['stdout'] = sp.DEVNULL
_old_call(*args, **kwargs)
sp.call = _call_nostderr
# Preallocating the pool of processes.
pool = ProcessPoolExecutor(4)
pool.__enter__()
def interrupt():
global INTERRUPTING
INTERRUPTING = True
class FileCleaner:
def __init__(self, file_lifetime: float = 3600):
self.file_lifetime = file_lifetime
self.files = []
def add(self, path: tp.Union[str, Path]):
self._cleanup()
self.files.append((time.time(), Path(path)))
def _cleanup(self):
now = time.time()
for time_added, path in list(self.files):
if now - time_added > self.file_lifetime:
if path.exists():
path.unlink()
self.files.pop(0)
else:
break
# 10 minutes
file_cleaner = FileCleaner(600)
def make_waveform(*args, **kwargs):
# Further remove some warnings.
be = time.time()
with warnings.catch_warnings():
warnings.simplefilter('ignore')
out = gr.make_waveform(*args, **kwargs)
print("Make a video took", time.time() - be)
return out
def load_model(version='melody'):
global MODEL, DEMUCS_MODEL
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if LOCAL:
if MODEL is None or MODEL.name != version:
print("Loading model", version)
# If gpu is not available, we'll use cpu.
MODEL = MusicGen.get_pretrained(version, device=device)
if DEMUCS_MODEL is None:
DEMUCS_MODEL = pretrained.get_model('htdemucs').to(device)
def connect_to_endpoint():
global client
client = Client("https://facebook-musicgen--44zzp.hf.space/")
def _do_predictions(texts, melodies, duration, progress=False, **gen_kwargs):
MODEL.set_generation_params(duration=duration, **gen_kwargs)
print("new batch", len(texts), texts, [None if m is None else (m[0], m[1].shape) for m in melodies])
be = time.time()
processed_melodies = []
target_sr = 32000
target_ac = 1
for melody in melodies:
if melody is None:
processed_melodies.append(None)
else:
sr, melody = melody[0], torch.from_numpy(melody[1]).to(MODEL.device).float().t()
if melody.dim() == 1:
melody = melody[None]
melody = melody[..., :int(sr * duration)]
melody = convert_audio(melody, sr, target_sr, target_ac)
processed_melodies.append(melody)
if any(m is not None for m in processed_melodies):
outputs = MODEL.generate_with_chroma(
descriptions=texts,
melody_wavs=processed_melodies,
melody_sample_rate=target_sr,
progress=progress,
)
else:
outputs = MODEL.generate(texts, progress=progress)
outputs = outputs.detach().float()
out_files = []
for output in outputs:
# Demucs
print("Running demucs")
wav = convert_audio(output, MODEL.sample_rate, DEMUCS_MODEL.samplerate, DEMUCS_MODEL.audio_channels)
wav = wav.unsqueeze(0)
stems = apply_model(DEMUCS_MODEL, wav)
stems = stems[:, stem_idx] # extract stem
stems = stems.sum(1) # merge extracted stems
stems = convert_audio(stems, DEMUCS_MODEL.samplerate, MODEL.sample_rate, 1)
demucs_output = stems[0]
output = output.cpu()
demucs_output = demucs_output.cpu()
# Naming
d_filename = f"temp/{texts[0][:10]}.wav"
# If path exists, add number. If number exists, update number.
i = 1
while Path(d_filename).exists():
d_filename = f"temp/{texts[0][:10]}_{i}.wav"
i += 1
audio_write(
d_filename, demucs_output, MODEL.sample_rate, strategy="loudness",
loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
out_files.append(d_filename)
file_cleaner.add(d_filename)
res = [out_file for out_file in out_files]
for file in res:
file_cleaner.add(file)
print("batch finished", len(texts), time.time() - be)
print("Tempfiles currently stored: ", len(file_cleaner.files))
return res
def predict_full(text, melody, progress=gr.Progress()):
global selected_text
global INTERRUPTING
INTERRUPTING = False
print("Running local model")
def _progress(generated, to_generate):
progress((generated, to_generate))
if INTERRUPTING:
raise gr.Error("Interrupted.")
MODEL.set_custom_progress_callback(_progress)
outs = _do_predictions(
[text], [melody], duration=10, progress=True)
selected_text = text
return outs[0], gr.File.update(value=outs[0], visible=True)
def select_new_melody():
global selected_melody
with NamedTemporaryFile("wb", suffix=".wav", delete=False) as file:
if not USE_MIDI:
new_melody_file = np.random.choice(melody_files)
selected_melody = new_melody_file
else:
new_melody_file = np.random.choice(midi_files)
selected_melody = new_melody_file
new_melody_file = render_midi(new_melody_file, fname=file.name)
crop_melody(new_melody_file, fname=file.name)
file_cleaner.add(file.name)
return file.name
def render_midi(midi_file, fname):
# sonify midi as sine wave
pm = pretty_midi.PrettyMIDI(midi_file)
sine_waves = pm.synthesize(fs=32000)
audio_write(fname, torch.from_numpy(sine_waves), 32000, strategy="loudness", loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
return fname
def crop_melody(melody_file, fname):
global selected_crop
crop = np.random.choice(len(crops))
crop = crops[crop]
selected_crop = crop
melody, sr = audio_read(melody_file)
melody = melody[:, crop[0]*sr:crop[1]*sr]
audio_write(fname, melody, sr, strategy="loudness", loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
def run_remote_model(text, melody):
global selected_text
print("Running Audiocraft API model with text", text, "and melody", melody.split("/")[-1])
result = client.predict(
text, # str in 'Describe your music' Textbox component
melody, # str (filepath or URL to file) in 'File' Audio component
fn_index=0
)
# Naming
d_filename = os.path.join("temp", f"{text[:10]}.wav")
# If path exists, add number. If number exists, update number.
i = 1
while Path(d_filename).exists():
d_filename = os.path.join("temp", f"{text[:10]}_{i}.wav")
i += 1
# Convert mp4 to wav, using ffmpeg
# ffmpeg -i input.mp4 -vn -acodec pcm_s16le -ar 44100 -ac 2 output.wav
sp.run(["ffmpeg", "-i", result, "-vn", "-acodec", "pcm_s16le", "-ar", "32000", "-ac", "1", d_filename])
# Load wav file
output, sr = audio_read(d_filename)
# Crop to 10 seconds
output = output[:, :10*sr]
# Demucs
print("Running demucs")
wav = convert_audio(output, sr, DEMUCS_MODEL.samplerate, DEMUCS_MODEL.audio_channels)
wav = wav.unsqueeze(0)
stems = apply_model(DEMUCS_MODEL, wav)
stems = stems[:, stem_idx] # extract stem
stems = stems.sum(1) # merge extracted stems
stems = convert_audio(stems, DEMUCS_MODEL.samplerate, 32000, 1)
demucs_output = stems[0]
output = output.cpu()
demucs_output = demucs_output.cpu()
audio_write(
d_filename, demucs_output, 32000, strategy="loudness",
loudness_headroom_db=16, loudness_compressor=True, add_suffix=False)
file_cleaner.add(d_filename)
selected_text = text
print("Finished", text)
print("Tempfiles currently stored: ", len(file_cleaner.files))
return d_filename#, gr.File.update(value=d_filename, visible=True)
def rating_callback(rating):
print("TEXT:", selected_text, "MELODY:" ,selected_melody, "CROP:", selected_crop, "RATING:", rating)
def ui_full(launch_kwargs):
with gr.Blocks() as interface:
gr.Markdown(
"""
# Soundsauce Melody Playground
"""
)
with gr.Row():
with gr.Column():
with gr.Row():
text = gr.Text(label="Input Text", interactive=True)
with gr.Column():
# previously, type="numpy"
if LOCAL:
audio_type="numpy"
else:
audio_type="filepath"
melody = gr.Audio(type=audio_type, label="File", source="upload",
interactive=True, elem_id="melody-input", value=select_new_melody())
new_melody = gr.Button("New Melody", interactive=True)
with gr.Row():
submit = gr.Button("Submit")
# Adapted from https://github.com/rkfg/audiocraft/blob/long/app.py, MIT license.
# _ = gr.Button("Interrupt").click(fn=interrupt, queue=False)
with gr.Column():
output_without_drum = gr.Audio(label="Output")
with gr.Row():
slider = gr.Slider(label="Rating", minimum=0, maximum=10, step=1, value=0, scale=2)
submit_button = gr.Button("Submit Rating", scale=1)
# file_download_no_drum = gr.File(label="Download", visible=False)
# gr.Markdown(
# """
# Note that the files will be deleted after 10 minutes, so make sure to download!
# """
# )
if LOCAL:
submit.click(predict_full,
inputs=[text, melody],
outputs=[output_without_drum])#, file_download_no_drum])
else:
submit.click(run_remote_model, inputs=[text, melody], outputs=[output_without_drum])#, file_download_no_drum])
new_melody.click(select_new_melody, outputs=[melody])
# Button callbacks
submit_button.click(rating_callback, inputs=[slider])
gr.Examples(
fn=predict_full,
examples=[
["Enchanting Flute Trills amidst Misty String Section"],
["Gliding Mellotron Strings over Vibrant Phrases"],
["Synth Brass Melody Floating over Airy Wind Chimes"],
["Rhythmic Acoustic Guitar Licks with Echoing Layers"],
["Whimsical Flute Flourishes in a Mystical Forest Glade"],
["Airy Piccolo Trills accompanied by Floating Harp Arpeggios"],
["Dreamy Harp Glissandos accompanied by Distant Celesta"],
["Hypnotic Synth Pads layered with Enigmatic Guitar Progressions"],
["Enchanting Kalimba Melodies atop Mystical Atmosphere"],
],
inputs=[text],
outputs=[output_without_drum]#, file_download_no_drum]
)
interface.queue().launch(**launch_kwargs)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
'--listen',
type=str,
default='0.0.0.0' if 'SPACE_ID' in os.environ else '127.0.0.1',
help='IP to listen on for connections to Gradio',
)
parser.add_argument("--local", action="store_true", help="Run locally instead of using API")
parser.add_argument("--midi", action="store_true", help="Render midi instead of wav")
args = parser.parse_args()
launch_kwargs = {}
launch_kwargs['server_name'] = args.listen
print("Using midi:", USE_MIDI)
# Load melody model
load_model()
if not LOCAL:
connect_to_endpoint()
if not os.path.exists("temp"):
os.mkdir("temp")
# Show the interface
ui_full(launch_kwargs)
|