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 GenerateMusic(input_title, input_num_tokens, input_prompt_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 = 4096 # Models seq len PAD_IDX = 2571 # 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 = 8, 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('Text_to_Music_Transformer_Medium_Trained_Model_33934_steps_0.6093_loss_0.813_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) input_num_tokens = max(8, min(2048, input_num_tokens)) print('-' * 70) print('Input title:', input_title) print('Req num toks:', input_num_tokens) print('Open-ended prompt:', input_prompt_type) print('-' * 70) #=============================================================================== print('Setting up model patches and loading helper functions...') # @title Setup and load model channels MIDI patches model_channel_0_piano_family = "Acoustic Grand" # @param ["Acoustic Grand", "Bright Acoustic", "Electric Grand", "Honky-Tonk", "Electric Piano 1", "Electric Piano 2", "Harpsichord", "Clav"] model_channel_1_chromatic_percussion_family = "Music Box" # @param ["Celesta", "Glockenspiel", "Music Box", "Vibraphone", "Marimba", "Xylophone", "Tubular Bells", "Dulcimer"] model_channel_2_organ_family = "Church Organ" # @param ["Drawbar Organ", "Percussive Organ", "Rock Organ", "Church Organ", "Reed Organ", "Accordion", "Harmonica", "Tango Accordion"] model_channel_3_guitar_family = "Acoustic Guitar(nylon)" # @param ["Acoustic Guitar(nylon)", "Acoustic Guitar(steel)", "Electric Guitar(jazz)", "Electric Guitar(clean)", "Electric Guitar(muted)", "Overdriven Guitar", "Distortion Guitar", "Guitar Harmonics"] model_channel_4_bass_family = "Fretless Bass" # @param ["Acoustic Bass", "Electric Bass(finger)", "Electric Bass(pick)", "Fretless Bass", "Slap Bass 1", "Slap Bass 2", "Synth Bass 1", "Synth Bass 2"] model_channel_5_strings_family = "Violin" # @param ["Violin", "Viola", "Cello", "Contrabass", "Tremolo Strings", "Pizzicato Strings", "Orchestral Harp", "Timpani"] model_channel_6_ensemble_family = "Choir Aahs" # @param ["String Ensemble 1", "String Ensemble 2", "SynthStrings 1", "SynthStrings 2", "Choir Aahs", "Voice Oohs", "Synth Voice", "Orchestra Hit"] model_channel_7_brass_family = "Trumpet" # @param ["Trumpet", "Trombone", "Tuba", "Muted Trumpet", "French Horn", "Brass Section", "SynthBrass 1", "SynthBrass 2"] model_channel_8_reed_family = "Alto Sax" # @param ["Soprano Sax", "Alto Sax", "Tenor Sax", "Baritone Sax", "Oboe", "English Horn", "Bassoon", "Clarinet"] model_channel_9_pipe_family = "Flute" # @param ["Piccolo", "Flute", "Recorder", "Pan Flute", "Blown Bottle", "Skakuhachi", "Whistle", "Ocarina"] model_channel_10_synth_lead_family = "Lead 8 (bass+lead)" # @param ["Lead 1 (square)", "Lead 2 (sawtooth)", "Lead 3 (calliope)", "Lead 4 (chiff)", "Lead 5 (charang)", "Lead 6 (voice)", "Lead 7 (fifths)", "Lead 8 (bass+lead)"] model_channel_11_synth_pad_family = "Pad 2 (warm)" # @param ["Pad 1 (new age)", "Pad 2 (warm)", "Pad 3 (polysynth)", "Pad 4 (choir)", "Pad 5 (bowed)", "Pad 6 (metallic)", "Pad 7 (halo)", "Pad 8 (sweep)"] model_channel_12_synth_effects_family = "FX 3 (crystal)" # @param ["FX 1 (rain)", "FX 2 (soundtrack)", "FX 3 (crystal)", "FX 4 (atmosphere)", "FX 5 (brightness)", "FX 6 (goblins)", "FX 7 (echoes)", "FX 8 (sci-fi)"] model_channel_13_ethnic_family = "Banjo" # @param ["Sitar", "Banjo", "Shamisen", "Koto", "Kalimba", "Bagpipe", "Fiddle", "Shanai"] model_channel_14_percussive_family = "Melodic Tom" # @param ["Tinkle Bell", "Agogo", "Steel Drums", "Woodblock", "Taiko Drum", "Melodic Tom", "Synth Drum", "Reverse Cymbal"] model_channel_15_sound_effects_family = "Bird Tweet" # @param ["Guitar Fret Noise", "Breath Noise", "Seashore", "Bird Tweet", "Telephone Ring", "Helicopter", "Applause", "Gunshot"] model_channel_16_drums_family = "Drums" # @param ["Drums"] print('=' * 70) print('Loading helper functions...') def txt2tokens(txt): return [ord(char)+2440 if 0 < ord(char) < 128 else 0+2440 for char in txt.lower()] def tokens2txt(tokens): return [chr(tok-2440) for tok in tokens if 0+2440 < tok < 128+2440 ] print('=' * 70) print('Setting up patches...') print('=' * 70) instruments = [v[1] for v in TMIDIX.Number2patch.items()] patches = [instruments.index(model_channel_0_piano_family), instruments.index(model_channel_1_chromatic_percussion_family), instruments.index(model_channel_2_organ_family), instruments.index(model_channel_3_guitar_family), instruments.index(model_channel_4_bass_family), instruments.index(model_channel_5_strings_family), instruments.index(model_channel_6_ensemble_family), instruments.index(model_channel_7_brass_family), instruments.index(model_channel_8_reed_family), 9, # Drums patch instruments.index(model_channel_9_pipe_family), instruments.index(model_channel_10_synth_lead_family), instruments.index(model_channel_11_synth_pad_family), instruments.index(model_channel_12_synth_effects_family), instruments.index(model_channel_13_ethnic_family), instruments.index(model_channel_15_sound_effects_family) ] print('Done!') print('=' * 70) print('Generating...') #@title Standard Text-to-Music Generator #@markdown Prompt settings song_title_prompt = input_title open_ended_prompt = input_prompt_type #@markdown Generation settings number_of_tokens_to_generate = input_num_tokens number_of_batches_to_generate = 1 #@param {type:"slider", min:1, max:16, step:1} temperature = 0.9 # @param {type:"slider", min:0.1, max:1, step:0.05} print('=' * 70) print('Text-to-Music Model Generator') print('=' * 70) if song_title_prompt == '': outy = [2569] else: if open_ended_prompt: outy = [2569] + txt2tokens(song_title_prompt) else: outy = [2569] + txt2tokens(song_title_prompt) + [2570] print('Selected prompt sequence:') print(outy[:12]) print('=' * 70) torch.cuda.empty_cache() inp = [outy] * number_of_batches_to_generate inp = torch.LongTensor(inp).cuda() with ctx: out = model.generate(inp, number_of_tokens_to_generate, temperature=temperature, return_prime=True, verbose=False) out0 = out.tolist() print('=' * 70) print('Done!') print('=' * 70) #=============================================================================== print('Rendering results...') print('=' * 70) out1 = out0[0] print('Sample INTs', out1[:12]) print('=' * 70) generated_song_title = ''.join(tokens2txt(out1)).title() print('Generated song title:', generated_song_title) print('=' * 70) if len(out1) != 0: song = out1 song_f = [] time = 0 dur = 0 vel = 90 pitch = 0 channel = 0 chan = 0 for ss in song: if 0 <= ss < 128: time += ss * 32 if 128 <= ss < 256: dur = (ss-128) * 32 if 256 <= ss < 2432: chan = (ss-256) // 128 if chan < 9: channel = chan elif 9 < chan < 15: channel = chan+1 elif chan == 15: channel = 15 elif chan == 16: channel = 9 pitch = (ss-256) % 128 if 2432 <= ss < 2440: vel = (((ss-2432)+1) * 15)-1 song_f.append(['note', time, dur, channel, pitch, vel, chan*8 ]) fn1 = "Text-to-Music-Transformer-Composition" detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(song_f, output_signature = 'Text-to-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 = generated_song_title 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:', 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("

Descriptive Music Transformer

") gr.Markdown("

A music transformer that describes music it generates

") gr.Markdown( "![Visitors](https://api.visitorbadge.io/api/visitors?path=asigalov61.Descriptive-Music-Transformer&style=flat)\n\n" "Generate music based on a title of your imagination :)\n\n" "Check out [Annotated MIDI Dataset](https://huggingface.co/datasets/asigalov61/Annotated-MIDI-Dataset) on Hugging Face!\n\n" "[Open In Colab]" "(https://colab.research.google.com/github/asigalov61/Text-to-Music-Transformer/blob/main/Text_to_Music_Transformer.ipynb)" " for faster execution and endless generation" ) gr.Markdown("## Enter any desired song title") input_title = gr.Textbox(value="Nothing Else Matters", label="Song title") input_prompt_type = gr.Checkbox(label="Open-ended prompt") input_num_tokens = gr.Slider(8, 2048, value=512, step=8, label="Number of tokens to generate") run_btn = gr.Button("generate", variant="primary") gr.Markdown("## Generation results") output_midi_title = gr.Textbox(label="Generated 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(GenerateMusic, [input_title, input_num_tokens, input_prompt_type], [output_midi_title, output_midi_summary, output_midi, output_audio, output_plot]) app.queue().launch()