Commit
•
a44bbef
1
Parent(s):
29230bd
Update app.py
Browse files
app.py
CHANGED
@@ -23,15 +23,16 @@ in_space = os.getenv("SYSTEM") == "spaces"
|
|
23 |
# =================================================================================================
|
24 |
|
25 |
@spaces.GPU
|
26 |
-
def GenerateMusic(input_title
|
|
|
27 |
print('=' * 70)
|
28 |
print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
|
29 |
start_time = reqtime.time()
|
30 |
|
31 |
print('Loading model...')
|
32 |
|
33 |
-
SEQ_LEN =
|
34 |
-
PAD_IDX =
|
35 |
DEVICE = 'cuda' # 'cuda'
|
36 |
|
37 |
# instantiate the model
|
@@ -39,10 +40,10 @@ def GenerateMusic(input_title, input_num_tokens, input_prompt_type):
|
|
39 |
model = TransformerWrapper(
|
40 |
num_tokens = PAD_IDX+1,
|
41 |
max_seq_len = SEQ_LEN,
|
42 |
-
attn_layers = Decoder(dim =
|
43 |
)
|
44 |
|
45 |
-
model = AutoregressiveWrapper(model, ignore_index = PAD_IDX)
|
46 |
|
47 |
model.to(DEVICE)
|
48 |
print('=' * 70)
|
@@ -50,7 +51,7 @@ def GenerateMusic(input_title, input_num_tokens, input_prompt_type):
|
|
50 |
print('Loading model checkpoint...')
|
51 |
|
52 |
model.load_state_dict(
|
53 |
-
torch.load('
|
54 |
map_location=DEVICE))
|
55 |
print('=' * 70)
|
56 |
|
@@ -66,83 +67,25 @@ def GenerateMusic(input_title, input_num_tokens, input_prompt_type):
|
|
66 |
print('Done!')
|
67 |
print('=' * 70)
|
68 |
|
69 |
-
input_num_tokens =
|
70 |
-
|
71 |
-
print('-' * 70)
|
72 |
-
print('Input title:', input_title)
|
73 |
-
print('Req num toks:', input_num_tokens)
|
74 |
-
print('Open-ended prompt:', input_prompt_type)
|
75 |
print('-' * 70)
|
76 |
|
77 |
#===============================================================================
|
78 |
|
79 |
-
print('Setting up model patches and loading helper functions...')
|
80 |
-
|
81 |
-
# @title Setup and load model channels MIDI patches
|
82 |
-
|
83 |
-
model_channel_0_piano_family = "Acoustic Grand" # @param ["Acoustic Grand", "Bright Acoustic", "Electric Grand", "Honky-Tonk", "Electric Piano 1", "Electric Piano 2", "Harpsichord", "Clav"]
|
84 |
-
model_channel_1_chromatic_percussion_family = "Music Box" # @param ["Celesta", "Glockenspiel", "Music Box", "Vibraphone", "Marimba", "Xylophone", "Tubular Bells", "Dulcimer"]
|
85 |
-
model_channel_2_organ_family = "Church Organ" # @param ["Drawbar Organ", "Percussive Organ", "Rock Organ", "Church Organ", "Reed Organ", "Accordion", "Harmonica", "Tango Accordion"]
|
86 |
-
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"]
|
87 |
-
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"]
|
88 |
-
model_channel_5_strings_family = "Violin" # @param ["Violin", "Viola", "Cello", "Contrabass", "Tremolo Strings", "Pizzicato Strings", "Orchestral Harp", "Timpani"]
|
89 |
-
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"]
|
90 |
-
model_channel_7_brass_family = "Trumpet" # @param ["Trumpet", "Trombone", "Tuba", "Muted Trumpet", "French Horn", "Brass Section", "SynthBrass 1", "SynthBrass 2"]
|
91 |
-
model_channel_8_reed_family = "Alto Sax" # @param ["Soprano Sax", "Alto Sax", "Tenor Sax", "Baritone Sax", "Oboe", "English Horn", "Bassoon", "Clarinet"]
|
92 |
-
model_channel_9_pipe_family = "Flute" # @param ["Piccolo", "Flute", "Recorder", "Pan Flute", "Blown Bottle", "Skakuhachi", "Whistle", "Ocarina"]
|
93 |
-
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)"]
|
94 |
-
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)"]
|
95 |
-
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)"]
|
96 |
-
model_channel_13_ethnic_family = "Banjo" # @param ["Sitar", "Banjo", "Shamisen", "Koto", "Kalimba", "Bagpipe", "Fiddle", "Shanai"]
|
97 |
-
model_channel_14_percussive_family = "Melodic Tom" # @param ["Tinkle Bell", "Agogo", "Steel Drums", "Woodblock", "Taiko Drum", "Melodic Tom", "Synth Drum", "Reverse Cymbal"]
|
98 |
-
model_channel_15_sound_effects_family = "Bird Tweet" # @param ["Guitar Fret Noise", "Breath Noise", "Seashore", "Bird Tweet", "Telephone Ring", "Helicopter", "Applause", "Gunshot"]
|
99 |
-
model_channel_16_drums_family = "Drums" # @param ["Drums"]
|
100 |
-
|
101 |
print('=' * 70)
|
102 |
print('Loading helper functions...')
|
103 |
|
104 |
def txt2tokens(txt):
|
105 |
-
return [ord(char)+
|
106 |
|
107 |
def tokens2txt(tokens):
|
108 |
-
return [chr(tok-
|
109 |
|
110 |
print('=' * 70)
|
111 |
-
print('Setting up patches...')
|
112 |
-
print('=' * 70)
|
113 |
-
|
114 |
-
instruments = [v[1] for v in TMIDIX.Number2patch.items()]
|
115 |
-
|
116 |
-
patches = [instruments.index(model_channel_0_piano_family),
|
117 |
-
instruments.index(model_channel_1_chromatic_percussion_family),
|
118 |
-
instruments.index(model_channel_2_organ_family),
|
119 |
-
instruments.index(model_channel_3_guitar_family),
|
120 |
-
instruments.index(model_channel_4_bass_family),
|
121 |
-
instruments.index(model_channel_5_strings_family),
|
122 |
-
instruments.index(model_channel_6_ensemble_family),
|
123 |
-
instruments.index(model_channel_7_brass_family),
|
124 |
-
instruments.index(model_channel_8_reed_family),
|
125 |
-
9, # Drums patch
|
126 |
-
instruments.index(model_channel_9_pipe_family),
|
127 |
-
instruments.index(model_channel_10_synth_lead_family),
|
128 |
-
instruments.index(model_channel_11_synth_pad_family),
|
129 |
-
instruments.index(model_channel_12_synth_effects_family),
|
130 |
-
instruments.index(model_channel_13_ethnic_family),
|
131 |
-
instruments.index(model_channel_15_sound_effects_family)
|
132 |
-
]
|
133 |
-
|
134 |
-
print('Done!')
|
135 |
-
print('=' * 70)
|
136 |
-
|
137 |
print('Generating...')
|
138 |
|
139 |
#@title Standard Text-to-Music Generator
|
140 |
|
141 |
-
#@markdown Prompt settings
|
142 |
-
|
143 |
-
song_title_prompt = input_title
|
144 |
-
open_ended_prompt = input_prompt_type
|
145 |
-
|
146 |
#@markdown Generation settings
|
147 |
|
148 |
number_of_tokens_to_generate = input_num_tokens
|
@@ -150,22 +93,10 @@ def GenerateMusic(input_title, input_num_tokens, input_prompt_type):
|
|
150 |
temperature = 0.9 # @param {type:"slider", min:0.1, max:1, step:0.05}
|
151 |
|
152 |
print('=' * 70)
|
153 |
-
print('
|
154 |
print('=' * 70)
|
155 |
|
156 |
-
|
157 |
-
outy = [2569]
|
158 |
-
|
159 |
-
else:
|
160 |
-
if open_ended_prompt:
|
161 |
-
outy = [2569] + txt2tokens(song_title_prompt)
|
162 |
-
|
163 |
-
else:
|
164 |
-
outy = [2569] + txt2tokens(song_title_prompt) + [2570]
|
165 |
-
|
166 |
-
print('Selected prompt sequence:')
|
167 |
-
print(outy[:12])
|
168 |
-
print('=' * 70)
|
169 |
|
170 |
torch.cuda.empty_cache()
|
171 |
|
@@ -313,25 +244,19 @@ if __name__ == "__main__":
|
|
313 |
"(https://colab.research.google.com/github/asigalov61/Text-to-Music-Transformer/blob/main/Text_to_Music_Transformer.ipynb)"
|
314 |
" for faster execution and endless generation"
|
315 |
)
|
316 |
-
gr.Markdown("## Enter any desired song title")
|
317 |
|
318 |
-
input_title = gr.Textbox(value="Nothing Else Matters", label="Song title")
|
319 |
-
input_prompt_type = gr.Checkbox(label="Open-ended prompt")
|
320 |
-
|
321 |
-
input_num_tokens = gr.Slider(8, 2048, value=512, step=8, label="Number of tokens to generate")
|
322 |
-
|
323 |
run_btn = gr.Button("generate", variant="primary")
|
324 |
|
325 |
gr.Markdown("## Generation results")
|
326 |
|
327 |
output_midi_title = gr.Textbox(label="Generated MIDI title")
|
328 |
-
output_midi_summary = gr.Textbox(label="
|
329 |
output_audio = gr.Audio(label="Output MIDI audio", format="wav", elem_id="midi_audio")
|
330 |
output_plot = gr.Plot(label="Output MIDI score plot")
|
331 |
output_midi = gr.File(label="Output MIDI file", file_types=[".mid"])
|
332 |
|
333 |
|
334 |
-
run_event = run_btn.click(GenerateMusic, [input_title
|
335 |
[output_midi_title, output_midi_summary, output_midi, output_audio, output_plot])
|
336 |
|
337 |
app.queue().launch()
|
|
|
23 |
# =================================================================================================
|
24 |
|
25 |
@spaces.GPU
|
26 |
+
def GenerateMusic(input_title):
|
27 |
+
|
28 |
print('=' * 70)
|
29 |
print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
|
30 |
start_time = reqtime.time()
|
31 |
|
32 |
print('Loading model...')
|
33 |
|
34 |
+
SEQ_LEN = 2048
|
35 |
+
PAD_IDX = 780
|
36 |
DEVICE = 'cuda' # 'cuda'
|
37 |
|
38 |
# instantiate the model
|
|
|
40 |
model = TransformerWrapper(
|
41 |
num_tokens = PAD_IDX+1,
|
42 |
max_seq_len = SEQ_LEN,
|
43 |
+
attn_layers = Decoder(dim = 1024, depth = 32, heads = 16, attn_flash = True)
|
44 |
)
|
45 |
|
46 |
+
model = AutoregressiveWrapper(model, ignore_index = PAD_IDX, pad_value=PAD_IDX)
|
47 |
|
48 |
model.to(DEVICE)
|
49 |
print('=' * 70)
|
|
|
51 |
print('Loading model checkpoint...')
|
52 |
|
53 |
model.load_state_dict(
|
54 |
+
torch.load('Descriptive_Music_Transformer_Trained_Model_20631_steps_0.3218_loss_0.8947_acc.pth',
|
55 |
map_location=DEVICE))
|
56 |
print('=' * 70)
|
57 |
|
|
|
67 |
print('Done!')
|
68 |
print('=' * 70)
|
69 |
|
70 |
+
input_num_tokens = 2040
|
|
|
|
|
|
|
|
|
|
|
71 |
print('-' * 70)
|
72 |
|
73 |
#===============================================================================
|
74 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
print('=' * 70)
|
76 |
print('Loading helper functions...')
|
77 |
|
78 |
def txt2tokens(txt):
|
79 |
+
return [ord(char)+648 if 0 < ord(char) < 128 else 0+648 for char in txt.lower()]
|
80 |
|
81 |
def tokens2txt(tokens):
|
82 |
+
return [chr(tok-648) for tok in tokens if 0+648 < tok < 128+648 ]
|
83 |
|
84 |
print('=' * 70)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
print('Generating...')
|
86 |
|
87 |
#@title Standard Text-to-Music Generator
|
88 |
|
|
|
|
|
|
|
|
|
|
|
89 |
#@markdown Generation settings
|
90 |
|
91 |
number_of_tokens_to_generate = input_num_tokens
|
|
|
93 |
temperature = 0.9 # @param {type:"slider", min:0.1, max:1, step:0.05}
|
94 |
|
95 |
print('=' * 70)
|
96 |
+
print('Descriptive Music Transformer Model Generator')
|
97 |
print('=' * 70)
|
98 |
|
99 |
+
outy = [777]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
|
101 |
torch.cuda.empty_cache()
|
102 |
|
|
|
244 |
"(https://colab.research.google.com/github/asigalov61/Text-to-Music-Transformer/blob/main/Text_to_Music_Transformer.ipynb)"
|
245 |
" for faster execution and endless generation"
|
246 |
)
|
|
|
247 |
|
|
|
|
|
|
|
|
|
|
|
248 |
run_btn = gr.Button("generate", variant="primary")
|
249 |
|
250 |
gr.Markdown("## Generation results")
|
251 |
|
252 |
output_midi_title = gr.Textbox(label="Generated MIDI title")
|
253 |
+
output_midi_summary = gr.Textbox(label="Generated music description")
|
254 |
output_audio = gr.Audio(label="Output MIDI audio", format="wav", elem_id="midi_audio")
|
255 |
output_plot = gr.Plot(label="Output MIDI score plot")
|
256 |
output_midi = gr.File(label="Output MIDI file", file_types=[".mid"])
|
257 |
|
258 |
|
259 |
+
run_event = run_btn.click(GenerateMusic, [input_title],
|
260 |
[output_midi_title, output_midi_summary, output_midi, output_audio, output_plot])
|
261 |
|
262 |
app.queue().launch()
|