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 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 GenerateAccompaniment(input_midi, input_num_tokens, input_acc_type):
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 = 8192 # Models seq len
PAD_IDX = 767 # 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 = 2048, depth = 4, 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('Ultimate_Accompaniment_Transformer_Small_Improved_Trained_Model_13649_steps_0.3229_loss_0.898_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)
fn = os.path.basename(input_midi.name)
fn1 = fn.split('.')[0]
input_num_tokens = max(4, min(128, input_num_tokens))
print('-' * 70)
print('Input file name:', fn)
print('Req num toks:', input_num_tokens)
print('Force acc:', input_acc_type)
print('-' * 70)
#===============================================================================
raw_score = TMIDIX.midi2single_track_ms_score(input_midi.name)
#===============================================================================
# Enhanced score notes
escore_notes = TMIDIX.advanced_score_processor(raw_score, return_enhanced_score_notes=True)[0]
escore_notes = [e for e in escore_notes if e[3] != 9]
if len(escore_notes) > 0:
#=======================================================
# PRE-PROCESSING
#===============================================================================
# Augmented enhanced score notes
escore_notes = TMIDIX.augment_enhanced_score_notes(escore_notes, timings_divider=32)
cscore = TMIDIX.chordify_score([1000, escore_notes])
melody = TMIDIX.fix_monophonic_score_durations([sorted(e, key=lambda x: x[4], reverse=True)[0] for e in cscore])
#=======================================================
# FINAL PROCESSING
melody_chords = []
#=======================================================
# MAIN PROCESSING CYCLE
#=======================================================
pe = cscore[0][0]
mpe = melody[0]
midx = 1
for i, c in enumerate(cscore):
c.sort(key=lambda x: (x[3], x[4]), reverse=True)
# Next melody note
if midx < len(melody):
# Time
mtime = melody[midx][1]-mpe[1]
mdur = melody[midx][2]
mdelta_time = max(0, min(127, mtime))
# Durations
mdur = max(0, min(127, mdur))
# Pitch
mptc = melody[midx][4]
else:
mtime = 127-mpe[1]
mdur = mpe[2]
mdelta_time = max(0, min(127, mtime))
# Durations
mdur = max(0, min(127, mdur))
# Pitch
mptc = mpe[4]
e = melody[i]
#=======================================================
# Timings...
time = e[1]-pe[1]
dur = e[2]
delta_time = max(0, min(127, time))
# Durations
dur = max(0, min(127, dur))
# Pitches
ptc = max(1, min(127, e[4]))
if ptc < 60:
ptc = 60 + (ptc % 12)
cha = e[3]
#=======================================================
# FINAL NOTE SEQ
if midx < len(melody):
melody_chords.append([delta_time, dur+128, ptc+384, mdelta_time+512, mptc+640])
mpe = melody[midx]
midx += 1
else:
melody_chords.append([delta_time, dur+128, ptc+384, mdelta_time+512, mptc+640])
pe = e
#===============================================================================
print('=' * 70)
print('Sample output events', melody_chords[:5])
print('=' * 70)
print('Generating...')
output = []
force_acc = input_acc_type
num_toks_per_note = 32
temperature=0.9
max_drums_limit=4
num_memory_tokens=4096
output1 = []
output2 = []
for m in melody_chords[:input_num_tokens]:
output1.extend(m)
input_seq = output1
if force_acc:
x = torch.LongTensor([input_seq+[0]]).cuda()
else:
x = torch.LongTensor([input_seq]).cuda()
time = input_seq[-2]-512
cur_time = 0
for _ in range(num_toks_per_note):
with ctx:
out = model.generate(x[-num_memory_tokens:],
1,
temperature=temperature,
return_prime=False,
verbose=False)
o = out.tolist()[0][0]
if 0 <= o < 128:
cur_time += o
if cur_time < time and o < 384:
out = torch.LongTensor([[o]]).cuda()
x = torch.cat((x, out), 1)
else:
break
outy = x.tolist()[0][len(input_seq):]
output1.extend(outy)
output2.append(outy)
print('=' * 70)
print('Done!')
print('=' * 70)
#===============================================================================
print('Rendering results...')
print('=' * 70)
print('Sample INTs', output1[:12])
print('=' * 70)
out1 = output2
accompaniment_MIDI_patch_number = 0
melody_MIDI_patch_number = 40
if len(out1) != 0:
song = out1
song_f = []
time = 0
ntime = 0
ndur = 0
vel = 90
npitch = 0
channel = 0
patches = [0] * 16
patches[0] = accompaniment_MIDI_patch_number
patches[3] = melody_MIDI_patch_number
for i, ss in enumerate(song):
ntime += melody_chords[i][0] * 32
ndur = (melody_chords[i][1]-128) * 32
nchannel = 1
npitch = (melody_chords[i][2]-256) % 128
vel = max(40, npitch)+20
song_f.append(['note', ntime, ndur, 3, npitch, vel, melody_MIDI_patch_number ])
time = ntime
for s in ss:
if 0 <= s < 128:
time += s * 32
if 128 <= s < 256:
dur = (s-128) * 32
if 256 <= s < 384:
pitch = (s-256)
vel = max(40, pitch)
song_f.append(['note', time, dur, 0, pitch, vel, accompaniment_MIDI_patch_number])
fn1 = "Ultimate-Accompaniment-Transformer-Composition"
detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(song_f,
output_signature = 'Ultimate Accompaniment 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("Ultimate Accompaniment Transformer
")
gr.Markdown("Generate unique accompaniment for any melody
")
gr.Markdown(
"\n\n"
"Accompaniment generation for any monophonic melody\n\n"
"Check out [Ultimate Drums Transformer](https://github.com/asigalov61/Ultimate-Accompaniment-Transformer) on GitHub!\n\n"
"[Open In Colab]"
"(https://colab.research.google.com/github/asigalov61/Ultimate-Accompaniment-Transformer/blob/main/Ultimate_Accompaniment_Transformer.ipynb)"
" for faster execution and endless generation"
)
gr.Markdown("## Upload your MIDI or select a sample example MIDI")
input_midi = gr.File(label="Input MIDI", file_types=[".midi", ".mid", ".kar"])
input_num_tokens = gr.Slider(4, 128, value=32, step=1, label="Number of composition chords to generate accompaniment for")
input_acc_type = gr.Checkbox(label='Force accompaniment generation for each melody note')
run_btn = gr.Button("generate", 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(GenerateAccompaniment, [input_midi, input_num_tokens, input_acc_type],
[output_midi_title, output_midi_summary, output_midi, output_audio, output_plot])
gr.Examples(
[["Ultimate-Accompaniment-Transformer-Melody-Seed-1.mid", 128, True],
["Ultimate-Accompaniment-Transformer-Melody-Seed-2.mid", 128, False],
["Ultimate-Accompaniment-Transformer-Melody-Seed-3.mid", 128, True],
["Ultimate-Accompaniment-Transformer-Melody-Seed-4.mid", 128, False],
["Ultimate-Accompaniment-Transformer-Melody-Seed-5.mid", 128, True],
["Ultimate-Accompaniment-Transformer-Melody-Seed-6.mid", 128, False],
["Ultimate-Accompaniment-Transformer-Melody-Seed-7.mid", 128, True]],
[input_midi, input_num_tokens, input_acc_type],
[output_midi_title, output_midi_summary, output_midi, output_audio, output_plot],
GenerateAccompaniment,
cache_examples=True,
)
app.queue().launch()