File size: 8,925 Bytes
bd71b6f
718967a
 
 
 
 
 
 
 
 
 
 
 
5d3e0a8
 
718967a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
77c722a
718967a
 
 
 
 
 
bd71b6f
718967a
 
 
 
bd71b6f
718967a
 
 
 
 
 
 
 
bd71b6f
718967a
6c0dab8
718967a
 
 
 
 
 
 
 
 
 
 
0e6d052
 
 
718967a
 
 
 
 
 
 
 
 
 
 
 
 
 
0e6d052
 
 
 
718967a
 
 
0e6d052
718967a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8bdfa67
 
718967a
 
 
 
 
 
 
77c722a
0e6d052
718967a
 
 
 
0e6d052
718967a
 
 
 
 
 
 
 
 
 
 
 
 
0e6d052
2508606
0e6d052
707c61d
718967a
 
 
 
 
 
2508606
718967a
 
707c61d
718967a
 
 
 
 
 
 
 
 
2508606
718967a
 
 
 
 
 
 
 
 
 
 
 
bbc9116
 
 
 
 
 
718967a
 
aff06f3
bd71b6f
718967a
 
 
 
23de607
718967a
 
 
 
2ef01c1
718967a
 
 
 
 
 
d3192fa
707c61d
718967a
d3192fa
 
 
 
718967a
77c722a
718967a
77c722a
718967a
77c722a
718967a
 
 
 
2508606
707c61d
 
8bdfa67
718967a
 
 
77c722a
2508606
718967a
 
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
# https://huggingface.co/spaces/asigalov61/LAKH-MIDI-Dataset-Search

import os

import time as reqtime
import datetime
from pytz import timezone

from sentence_transformers import SentenceTransformer
from sentence_transformers import util

import numpy as np

from datasets import load_dataset

import gradio as gr

import copy
import random
import pickle

import zlib

from midi_to_colab_audio import midi_to_colab_audio

import TMIDIX

import matplotlib.pyplot as plt

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

def find_midi(input_search_string):

    print('=' * 70)
    print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
    start_time = reqtime.time()

    print('-' * 70)
    print('Req search str:', input_search_string)
    print('-' * 70)
    
    print('Searching...')

    query_embedding = model.encode([input_search_string])
    
    # Compute cosine similarity between query and each sentence in the corpus
    similarities = util.cos_sim(query_embedding, corpus_embeddings)

    top_ten_matches_idxs = np.argsort(-similarities)[0][:10].tolist()
   
    # Find the index of the most similar sentence
    closest_index = np.argmax(similarities)
    closest_index_match_ratio = max(similarities[0].tolist())

    best_corpus_match = mc_dataset['train'][closest_index.tolist()]

    print('Done!')
    print('=' * 70)
    
    print('Match corpus index', closest_index)
    print('Match corpus ratio', closest_index_match_ratio)
    
    print('=' * 70)
    print('Done!')
    print('=' * 70)

    LAKH_hash = best_corpus_match['location'].split('/')[-1].split('.mid')[0]
    LAKH_caption = str(best_corpus_match['caption'])
    zlib_file_name = all_MIDI_files_names[MIDI_files_names.index(LAKH_hash)][1]

    print('Fetching MIDI score...')

    with open(zlib_file_name, 'rb') as f:
        compressed_data = f.read()
    
    # Decompress the data
    decompressed_data = zlib.decompress(compressed_data)
    
    # Convert the bytes back to a list using pickle
    scores_data = pickle.loads(decompressed_data)
    
    fnames = [f[0] for f in scores_data]
    
    fnameidx = fnames.index(LAKH_hash)

    MIDI_score_metadata = scores_data[fnameidx][1]
    MIDI_score_data = scores_data[fnameidx][2]
    
    print('Rendering results...')
    print('=' * 70)
    print('MIDi Title:', LAKH_hash)
    print('Sample INTs', MIDI_score_data[:12])
    print('=' * 70)
    
    if len(MIDI_score_data) != 0:
    
        song = MIDI_score_data
        song_f = []
        
        time = 0
        dur = 0
        vel = 90
        pitch = 0
        channel = 0
        
        patches = [-1] * 16
        
        channels = [0] * 16
        channels[9] = 1
        
        for ss in song:
    
            if 0 <= ss < 256:
            
              time += ss * 16
            
            if 256 <= ss < 512:
            
              dur = (ss-256) * 16
            
            if 512 <= ss <= 640:

                patch = (ss-512)
                
                if patch < 128:
            
                  if patch not in patches:
                    if 0 in channels:
                        cha = channels.index(0)
                        channels[cha] = 1
                    else:
                        cha = 15
            
                    patches[cha] = patch
                    channel = patches.index(patch)
                  else:
                    channel = patches.index(patch)
            
                if patch == 128:
                  channel = 9
                
            if 640 < ss < 768:

                ptc = (ss-640)

            if 768 < ss < 896:

                vel = (ss - 768)
            
                song_f.append(['note', time, dur, channel, ptc, vel, patch ])
    
    patches = [0 if x==-1 else x for x in patches]

    song_f = song_f[:3000]

    print('=' * 70)
    
    #===============================================================================
    
    output_score, patches, overflow_patches = TMIDIX.patch_enhanced_score_notes(song_f)
    
    detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(output_score,
                                                          output_signature = 'LAKH MIDI Dataset Search',
                                                          output_file_name = LAKH_hash,
                                                          track_name='Project Los Angeles',
                                                          list_of_MIDI_patches=patches
                                                          )
    
    new_fn = LAKH_hash + '.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(LAKH_hash)
    output_midi_caption = str(best_corpus_match['caption'])
    output_midi_summary = str(MIDI_score_metadata)
    output_midi_caps = str(best_corpus_match)
    output_midi = str(new_fn)
    output_audio = (16000, audio)
    
    output_plot = TMIDIX.plot_ms_SONG(output_score, plot_title=output_midi_title, return_plt=True)
    
    print('Output MIDI file name:', output_midi)
    print('Output MIDI caption string:', output_midi_caption)
    print('Output MIDI title:', output_midi_title)
    print('Output MIDI summary:', output_midi_summary)
    print('Output MidiCaps information:', output_midi_caps)
    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_caption, output_midi_summary, output_midi_caps, 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"

    print('Loading MidiCaps dataset...')
    
    mc_dataset = load_dataset("amaai-lab/MidiCaps")
    print('=' * 70)
    
    print('Loading files list...')
    
    all_MIDI_files_names = TMIDIX.Tegridy_Any_Pickle_File_Reader('LAKH_all_files_names')
    MIDI_files_names = [f[0] for f in all_MIDI_files_names]
    print('=' * 70)

    print('Loading MIDI corpus embeddings...')
    
    corpus_embeddings = np.load('MIDI_corpus_embeddings_all-mpnet-base-v2.npz.gz.npz')['data']
    print('Done!')
    print('=' * 70)

    print('Loading Sentence Transformer model...')
    model = SentenceTransformer('all-mpnet-base-v2')
    print('Done!')
    print('=' * 70)
    
    app = gr.Blocks()
    
    with app:
        gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>LAKH MIDI Dataset Search</h1>")
        gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>Search and explore LAKH MIDI dataset with MidiCaps dataset and sentence transformer</h1>")
        
        gr.Markdown("![Visitors](https://api.visitorbadge.io/api/visitors?path=asigalov61.LAKH-MIDI-Dataset-Search&style=flat)\n\n"
                    "This is a demo for MidiCaps dataset\n\n"
                    "Check out [MidiCaps Dataset](https://huggingface.co/datasets/amaai-lab/MidiCaps) on Hugging Face!\n\n"
                   )

        gr.Markdown("# Enter any desired song description\n\n")
        
        input_search_string = gr.Textbox(label="Search string", value="Cheery pop song about love and happiness")
        submit = gr.Button(value='Search')
        gr.ClearButton(components=[input_search_string])

        gr.Markdown("# Search results")

        output_midi_title = gr.Textbox(label="Output MIDI title")
        output_midi_caption = gr.Textbox(label="MIDI caption string")
        output_midi_summary = gr.Textbox(label="Aggregated MIDI file text metadata")
        output_midi_caps = gr.Textbox(label="MidiCaps dataset information")
        output_audio = gr.Audio(label="Output MIDI audio", format="mp3", 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 = submit.click(find_midi, [input_search_string],
                                                  [output_midi_title, output_midi_caption, output_midi_summary, output_midi_caps, output_midi, output_audio, output_plot ])
        
    app.launch()