app.py

#==================================================================================
# https://huggingface.co/spaces/asigalov61/Ultimate-Chords-Progressions-Transformer
#==================================================================================

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 statistics
import copy

import tqdm

from midi_to_colab_audio import midi_to_colab_audio
import TMIDIX

import matplotlib.pyplot as plt

# =================================================================================================
                       
@spaces.GPU
def Generate_Chords(input_midi, input_num_prime_chords, input_num_gen_chords):
    print('=' * 70)
    print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
    start_time = reqtime.time()
    
    print('=' * 70)
    print('Instantiating the model...')

    SEQ_LEN = 8192
    PAD_IDX = 2239
    DEVICE = 'cuda' # 'cpu'

    # instantiate the model

    model = TransformerWrapper(
        num_tokens = PAD_IDX+1,
        max_seq_len = SEQ_LEN,
        attn_layers = Decoder(dim = 2048, 
                              depth = 8, 
                              heads = 32, 
                              rotary_pos_emb = True,  
                              attn_flash = True
                             )
        )
    
    model = AutoregressiveWrapper(model, ignore_index = PAD_IDX, pad_value=PAD_IDX)

    model.to(DEVICE)
    
    print('Done!')
    print('=' * 70)

    print('Loading model checkpoint...')

    model.load_state_dict(
        torch.load('Ultimate_Chords_Progressions_Transformer_Trained_Model_LAX_5858_steps_0.4506_loss_0.8724_acc.pth',
                   map_location=DEVICE))

    model.eval()
    
    print('Done!')
    print('=' * 70)


    if DEVICE == 'cpu':
        dtype = torch.bfloat16
    else:
        dtype = torch.bfloat16

    ctx = torch.amp.autocast(device_type=DEVICE, dtype=dtype)

    print('Done!')
    print('=' * 70)

    fn = os.path.basename(input_midi.name)
    fn1 = fn.split('.')[0]

    print('=' * 70)
    print('Input file name:', fn)
    print('Num prime chords:', input_num_prime_chords)
    print('Num gen chords:', input_num_gen_chords)
    print('=' * 70)

    #===============================================================================
    
    raw_score = TMIDIX.midi2single_track_ms_score(input_midi.name)
    
    escore_notes = TMIDIX.advanced_score_processor(raw_score, return_enhanced_score_notes=True)
    
    if escore_notes:
        
        escore_notes = TMIDIX.augment_enhanced_score_notes(escore_notes[0], timings_divider=32, legacy_timings=True)
    
        if escore_notes:
    
            #=======================================================
            # PRE-PROCESSING
    
            # checking number of instruments in a composition
            instruments_list = sorted(set([e[6] for e in escore_notes]))
            instruments_list_without_drums = sorted(set([e[6] for e in escore_notes if e[3] != 9]))
            main_instruments_list = sorted(set([e[6] for e in escore_notes if e[6] < 80]))
            
            comp_times = [e[1] for e in escore_notes if e[6] < 80]
                
            comp_dtimes = [max(1, min(127, b-a)) for a, b in zip(comp_times[:-1], comp_times[1:]) if b-a != 0]
            avg_comp_dtime = max(0, min(127, int(sum(comp_dtimes) / len(comp_dtimes))))
            
            #=======================================================
            # FINAL PROCESSING

            #=======================================================
            # Adjusting avg velocity

            vels = [e[5] for e in escore_notes]
            avg_vel = int(sum(vels) / len(vels))

            if avg_vel < 60:
                TMIDIX.adjust_score_velocities(escore_notes, avg_vel * 2)

            melody_chords = []
            melody_chords2 = []
            mel_cho = []

            #=======================================================
            # Break between compositions / Intro seq

            if 128 in instruments_list:
                drums_present = 1931 # Yes
            else:
                drums_present = 1930 # No

            melody_chords.extend([1929, drums_present])
            mel_cho.extend([1929, drums_present])

            #=======================================================
            # Composition patches list

            melody_chords.extend([i+1932 for i in instruments_list_without_drums])
            mel_cho.extend([i+1932 for i in instruments_list_without_drums])
            #=======================================================
            # Composition avg pitch and dtime

            mode_instruments_pitch = statistics.mode([e[4] for e in escore_notes if e[6] < 80])

            melody_chords.extend([2060+mode_instruments_pitch, 2188+avg_comp_dtime])
            mel_cho.extend([2060+mode_instruments_pitch, 2188+avg_comp_dtime])

            melody_chords2.append(mel_cho)

            #=======================================================
            # MAIN PROCESSING CYCLE
            #=======================================================

            cscore = TMIDIX.chordify_score([1000, escore_notes])

            pc = cscore[0] # Previous chord

            for i, c in enumerate(cscore):

                c.sort(key=lambda x: x[6]) # Sorting by patch

                #=======================================================
                # Outro seq

                #if len(cscore) > 256:
                #    if len(cscore) - i == 64:
                #        melody_chords.extend([2236])

                #=======================================================
                # Timings...

                # Cliping all values...
                delta_time = max(0, min(127, c[0][1]-pc[0][1]))

                #=======================================================
                # Chords...

                cpitches = sorted([e[4] for e in c if e[3] != 9])
                dpitches = [e[4] for e in c if e[3] == 9]

                tones_chord = sorted(set([p % 12 for p in cpitches]))

                if tones_chord:

                    if tones_chord not in TMIDIX.ALL_CHORDS_SORTED:
                        tones_chord_tok = 644
                        tones_chord_tok = TMIDIX.ALL_CHORDS_SORTED.index(TMIDIX.advanced_check_and_fix_tones_chord(tones_chord, cpitches[-1]))

                    else:
                        tones_chord_tok = TMIDIX.ALL_CHORDS_SORTED.index(tones_chord) # 321

                    if dpitches:

                        if tones_chord_tok == 644:
                            tones_chord_tok = 645
                        else:
                            tones_chord_tok += 321

                else:
                    tones_chord_tok = 643 # Drums-only chord

                #=======================================================
                # Writing chord/time...

                melody_chords.extend([tones_chord_tok, delta_time+646])
                
                mel_cho = []
                mel_cho.extend([tones_chord_tok, delta_time+646])

                #=======================================================
                # Notes...

                pp = -1

                for e in c:

                    #=======================================================
                    # Duration
                    dur = max(0, min(63, int(max(0, e[2] // 4) * 2)))

                    # Pitch
                    ptc = max(1, min(127, e[4]))

                    # Octo-velocity
                    vel = max(8, min(127, (max(1, e[5] // 8) * 8)))
                    velocity = round(vel / 15)-1

                    # Patch
                    pat = max(0, min(128, e[6]))
                    
                    if 7 < pat < 80:
                        ptc += 128

                    elif 79 < pat < 128:
                        ptc += 256

                    elif pat == 128:
                        ptc += 384

                    #=======================================================
                    # FINAL NOTE SEQ

                    # Writing final note asynchronously

                    dur_vel = (8 * dur) + velocity # 512

                    if pat != pp:
                        melody_chords.extend([pat+774, ptc+904, dur_vel+1416]) # 1928
                        mel_cho.extend([pat+774, ptc+904, dur_vel+1416])

                    else:
                        melody_chords.extend([ptc+904, dur_vel+1416])
                        mel_cho.extend([ptc+904, dur_vel+1416])

                    pp = pat

                pc = c

                melody_chords2.append(mel_cho)

                #=======================================================

            #melody_chords.extend([2237]) # EOS

            #=======================================================
            # TOTAL DICTIONARY SIZE 2237+1=2238
            #=======================================================
    
    print('Done!')
    print('=' * 70)
    print('Melody chords length:', len(melody_chords))
    print('=' * 70)
        
    #==================================================================

    print('=' * 70)
    
    print('Sample output events', melody_chords[:12])
    print('=' * 70)
    print('Generating...')

    output = []
    
    for m in melody_chords2[:input_num_prime_chords]:
        output.extend(m)
    
    for ct in tqdm.tqdm(melody_chords2[input_num_prime_chords:input_num_prime_chords+input_num_gen_chords]):

        output.extend(ct[:2])
  
        y = 774
    
        while y > 773:
    
            x = torch.LongTensor(output).to(DEVICE)
            
            with ctx:
                out = model.generate(x,
                                     1,
                                     filter_logits_fn=top_p,
                                     filter_kwargs={'thres': 0.96},
                                     temperature=0.9,
                                     return_prime=False,
                                     verbose=False)
            
            y = out.tolist()[0][0]
    
            if y > 773:
                output.append(y)

    print('=' * 70)
    print('Done!')
    print('=' * 70)
    
    #===============================================================================
    print('Rendering results...')
    
    print('=' * 70)
    print('Sample INTs', output[:12])
    print('=' * 70)
    
    if len(output) != 0:
    
        song = output
        song_f = []
    
        time = 0
        dur = 4
        vel = 90
        pitch = 60
        channel = 0
        
        patches = [0] * 16
        patches[9] = 9
    
        for ss in song:
    
            if 645 < ss < 774:
    
                time += (ss-646)
                
            if 773 < ss < 904:
                
                pat = (ss - 774)
                     
                chan = (pat // 8)
                
                if 0 <= chan < 9:
                    channel = chan
                
                elif 8 < chan < 15:
                    channel = chan + 1
    
                elif chan == 16:
                    channel = 9
    
            if 903 < ss < 1416:
    
                pitch = (ss-904) % 128
    
            if 1415 < ss < 1928:
    
                dur = (((ss-1416) // 8)+1) * 2
                vel = (((ss-1416) % 8)+1) * 15
                
                song_f.append(['note', time, dur, channel, pitch, vel, pat])
    
    song_f, patches, overflow_patches = TMIDIX.patch_enhanced_score_notes(song_f)

    fn1 = "Ultimate-Chords-Progressions-Transformer-Composition"
    
    detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(song_f,
                                                              output_signature = 'Ultimate Chords Progressions Transformer',
                                                              output_file_name = fn1,
                                                              track_name='Project Los Angeles',
                                                              list_of_MIDI_patches=patches,
                                                              timings_multiplier=32
                                                              )
    
    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, timings_multiplier=32)

    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')
    print('*' * 70)

    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("<h1 style='text-align: center; margin-bottom: 1rem'>Ultimate Chords Progressions Transformer</h1>")
        gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>Self-correcting multi-instrumental chords-conditioned music RoPE transformer</h1>")
        gr.Markdown(
            "![Visitors](https://api.visitorbadge.io/api/visitors?path=asigalov61.Ultimate-Chords-Progressions-Transformer&style=flat)\n\n"
            "Check out [Ultimate Chords Progressions Transformer](https://huggingface.co/asigalov61/Ultimate-Chords-Progressions-Transformer) on Hugging Face!\n\n"
            "[Open In Colab]"
            "(https://colab.research.google.com/github/asigalov61/Chords-Progressions-Transformer/blob/main/Chords_Progressions_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_prime_chords = gr.Slider(1, 128, value=32, step=1, label="Number of prime chords")
        input_num_gen_chords = gr.Slider(4, 256, value=128, step=1, label="Number of composition chords to generate progression for")
        
        run_btn = gr.Button("Generate Chords", 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(Generate_Chords, [input_midi, input_num_prime_chords, input_num_gen_chords],
                                  [output_midi_title, output_midi_summary, output_midi, output_audio, output_plot])

        gr.Examples(
            [["Chords-Progressions-Transformer-MI-Seed-1.mid", 32, 128],
             ["Chords-Progressions-Transformer-MI-Seed-2.mid", 32, 128],
             ["Chords-Progressions-Transformer-MI-Seed-3.mid", 32, 128],
             ["Chords-Progressions-Transformer-MI-Seed-4.mid", 32, 128],
             ["Chords-Progressions-Transformer-MI-Seed-5.mid", 32, 128],
             ["Chords-Progressions-Transformer-MI-Seed-6.mid", 32, 128]
            ],
            [input_midi, input_num_prime_chords, input_num_gen_chords],
            [output_midi_title, output_midi_summary, output_midi, output_audio, output_plot],
            Generate_Chords,
            cache_examples=True,
            cache_mode='eager'
        )
        
        app.queue().launch()