import os.path
import time as reqtime
import datetime
from pytz import timezone
import torch
import spaces
import gradio as gr
from x_transformer_1_23_2 import *
import random
import copy
import tqdm
from midi_to_colab_audio import midi_to_colab_audio
import TMIDIX
import matplotlib.pyplot as plt
in_space = os.getenv("SYSTEM") == "spaces"
# =================================================================================================
@spaces.GPU
def GenerateGroove():
print('=' * 70)
print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
start_time = reqtime.time()
print('Loading model...')
SEQ_LEN = 4096 # Models seq len
PAD_IDX = 1664 # Models pad index
DEVICE = 'cuda' # 'cuda'
# instantiate the model
model = TransformerWrapper(
num_tokens = PAD_IDX+1,
max_seq_len = SEQ_LEN,
attn_layers = Decoder(dim = 1024, depth = 24, heads = 16, attn_flash = True)
)
model = AutoregressiveWrapper(model, ignore_index = PAD_IDX)
model.to(DEVICE)
print('=' * 70)
print('Loading model checkpoint...')
model.load_state_dict(
torch.load('Groove_Music_Transformer_Medium_Trained_Model_23268_steps_0.7459_loss_0.797_acc.pth',
map_location=DEVICE))
print('=' * 70)
model.eval()
if DEVICE == 'cpu':
dtype = torch.bfloat16
else:
dtype = torch.float16
ctx = torch.amp.autocast(device_type=DEVICE, dtype=dtype)
print('Done!')
print('=' * 70)
print('Loading Google Magenta Groove processed MIDIs...')
all_scores = TMIDIX.Tegridy_Any_Pickle_File_Reader('Google_Magenta_Groove_8675_Select_Processed_MIDIs')
print('Done!')
print('=' * 70)
print('=' * 70)
drums_score_idx = random.randint(0, len(all_scores))
drums_score_fn = all_scores[drums_score_idx][0]
drums_score = all_scores[drums_score_idx][1][:160]
print('Drums score index', drums_score_idx)
print('Drums score name', drums_score_fn)
print('Drums score length', len(drums_score))
print('=' * 70)
#==================================================================
print('=' * 70)
print('Sample input events', drums_score[:5])
print('=' * 70)
print('Prepping drums track...')
num_prime_chords = 7
outy = []
for d in drums_score[:num_prime_chords]:
outy.extend(d)
print('Generating...')
max_notes_per_chord=8
num_samples=4
num_memory_tokens = 4096
temperature=1.0
for i in range(num_prime_chords, len(drums_score)):
outy.extend(drums_score[i])
if i == num_prime_chords:
outy.append(256+12)
input_seq = outy[-num_memory_tokens:]
seq = copy.deepcopy(input_seq)
batch_value = 256
nc = 0
while batch_value > 255 and nc < max_notes_per_chord:
x = torch.tensor([seq] * num_samples, dtype=torch.long, device='cuda')
with ctx:
out = model.generate(x,
1,
temperature=temperature,
return_prime=False,
verbose=False)
out1 = [o[0] for o in out.tolist() if o[0] > 255]
if not out1:
out1 = [-1]
batch_value = random.choice(out1)
if batch_value > 255:
seq.append(batch_value)
if batch_value > 383:
nc += 1
out = seq[len(input_seq):]
outy.extend(out)
print('=' * 70)
print('Done!')
print('=' * 70)
#===============================================================================
print('Rendering results...')
print('=' * 70)
print('Sample INTs', outy[:12])
print('=' * 70)
if len(outy) != 0:
song = outy
song_f = []
time = 0
dur = 32
vel = 90
dvels = [100, 120]
pitch = 60
channel = 0
patches = [0, 10, 19, 24, 35, 40, 52, 56, 65, 9, 73, 0, 0, 0, 0, 0]
for ss in song:
if 0 <= ss < 128:
time += ss * 32
if 128 <= ss < 256:
song_f.append(['note', time, 32, 9, ss-128, dvels[(ss-128) % 2], 128])
if 256 < ss < 384:
dur = (ss-256) * 32
if 384 < ss < 1664:
chan = (ss-384) // 128
if chan == 11:
channel = 9
else:
if chan > 8:
channel = chan + 1
else:
channel = chan
if channel == 9:
patch = 128
else:
patch = channel * 8
pitch = (ss-384) % 128
vel = max(50, pitch)
song_f.append(['note', time, dur, channel, pitch, vel, patch])
fn1 = drums_score_fn
detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(song_f,
output_signature = 'Groove Music Transformer',
output_file_name = fn1,
track_name='Project Los Angeles',
list_of_MIDI_patches=patches
)
new_fn = fn1+'.mid'
audio = midi_to_colab_audio(new_fn,
soundfont_path=soundfont,
sample_rate=16000,
volume_scale=10,
output_for_gradio=True
)
print('Done!')
print('=' * 70)
#========================================================
output_midi_title = str(fn1)
output_midi_summary = str(song_f[:3])
output_midi = str(new_fn)
output_audio = (16000, audio)
output_plot = TMIDIX.plot_ms_SONG(song_f, plot_title=output_midi, return_plt=True)
print('Output MIDI file name:', output_midi)
print('Output MIDI title:', output_midi_title)
print('Output MIDI summary:', '')
print('=' * 70)
#========================================================
print('-' * 70)
print('Req end time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
print('-' * 70)
print('Req execution time:', (reqtime.time() - start_time), 'sec')
return output_midi_title, output_midi_summary, output_midi, output_audio, output_plot
# =================================================================================================
if __name__ == "__main__":
PDT = timezone('US/Pacific')
print('=' * 70)
print('App start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
print('=' * 70)
soundfont = "SGM-v2.01-YamahaGrand-Guit-Bass-v2.7.sf2"
app = gr.Blocks()
with app:
gr.Markdown("Groove Music Transformer
")
gr.Markdown("Generate music for Google Magenta Groove MIDI dataset drums tracks
")
gr.Markdown(
"\n\n"
"Generate music for Google Magenta Groove MIDI dataset drums tracks\n\n"
"Based upon [Google Magenta Groove MIDI Dataset](https://magenta.tensorflow.org/datasets/groove)\n\n"
)
run_btn = gr.Button("generate groove", variant="primary")
gr.Markdown("## Generation results")
output_midi_title = gr.Textbox(label="Output MIDI title")
output_midi_summary = gr.Textbox(label="Output MIDI summary")
output_audio = gr.Audio(label="Output MIDI audio", format="wav", elem_id="midi_audio")
output_plot = gr.Plot(label="Output MIDI score plot")
output_midi = gr.File(label="Output MIDI file", file_types=[".mid"])
run_event = run_btn.click(GenerateGroove,
outputs=[output_midi_title, output_midi_summary, output_midi, output_audio, output_plot])
app.queue().launch()