Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,25 +2,19 @@ import random
|
|
| 2 |
import argparse
|
| 3 |
import os
|
| 4 |
import glob
|
|
|
|
| 5 |
import rtmidi
|
| 6 |
import gradio as gr
|
| 7 |
import numpy as np
|
| 8 |
-
import
|
| 9 |
-
import base64
|
| 10 |
-
import io
|
| 11 |
-
import soundfile as sf # Placeholder for audio rendering
|
| 12 |
from huggingface_hub import hf_hub_download
|
|
|
|
| 13 |
from midi_synthesizer import MidiSynthesizer
|
|
|
|
| 14 |
|
| 15 |
MAX_SEED = np.iinfo(np.int32).max
|
| 16 |
in_space = os.getenv("SYSTEM") == "spaces"
|
| 17 |
|
| 18 |
-
SONG_DATA = {
|
| 19 |
-
"title": "Do You Believe in Love",
|
| 20 |
-
"progression": ["G", "D", "Em", "C"],
|
| 21 |
-
"lyrics": ["I was walking down a one-way street", "Just a-looking for someone to meet"]
|
| 22 |
-
}
|
| 23 |
-
|
| 24 |
class MIDIDeviceManager:
|
| 25 |
def __init__(self):
|
| 26 |
self.midiout = rtmidi.MidiOut()
|
|
@@ -51,63 +45,24 @@ class MIDIManager:
|
|
| 51 |
def __init__(self):
|
| 52 |
self.soundfont_path = hf_hub_download(repo_id="skytnt/midi-model", filename="soundfont.sf2")
|
| 53 |
self.synthesizer = MidiSynthesizer(self.soundfont_path)
|
| 54 |
-
self.loaded_midi = {}
|
| 55 |
-
self.modified_files = []
|
| 56 |
-
self.example_variations = {} # midi_id: (midi_base64, audio_base64) for pre-generated examples
|
| 57 |
self.is_playing = False
|
| 58 |
-
self.
|
| 59 |
-
self.
|
| 60 |
-
self.
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
def create_drum_beat(self):
|
| 70 |
-
return [(36, 100, 0), (42, 80, 50), (38, 90, 100), (42, 80, 150)] # Kick, hi-hat, snare, hi-hat
|
| 71 |
-
|
| 72 |
-
def create_starter_midi(self):
|
| 73 |
-
midi = MIDI.MIDIFile(5)
|
| 74 |
-
for i, inst in enumerate(self.instruments):
|
| 75 |
-
midi.addTrack()
|
| 76 |
-
midi.addProgramChange(i, 0, 0, inst)
|
| 77 |
-
for t in range(0, 400, 100):
|
| 78 |
-
note = random.randint(60, 84)
|
| 79 |
-
midi.addNote(i, 0, note, t, 100, 100)
|
| 80 |
-
midi.addTrack()
|
| 81 |
-
for note, vel, time in self.drum_beat:
|
| 82 |
-
midi.addNote(4, 9, note, time, 100, vel)
|
| 83 |
-
return midi
|
| 84 |
-
|
| 85 |
-
def preload_default_midi(self):
|
| 86 |
-
default_path = "default.mid"
|
| 87 |
-
if os.path.exists(default_path):
|
| 88 |
-
midi_id = "default"
|
| 89 |
-
midi = MIDI.load(default_path)
|
| 90 |
-
self.loaded_midi[midi_id] = (default_path, midi)
|
| 91 |
-
midi_data, audio_data = self.generate_variation(midi_id)
|
| 92 |
-
self.play_with_loop(midi_data)
|
| 93 |
-
|
| 94 |
-
def load_example_midis(self):
|
| 95 |
-
examples = {}
|
| 96 |
-
for file_path in glob.glob("*.mid") + glob.glob("*.midi"):
|
| 97 |
-
if file_path == "default.mid":
|
| 98 |
-
continue
|
| 99 |
-
midi_id = f"example_{len(examples)}"
|
| 100 |
-
midi = MIDI.load(file_path)
|
| 101 |
-
self.loaded_midi[midi_id] = (file_path, midi)
|
| 102 |
-
# Pre-generate variation for each example
|
| 103 |
-
midi_data, audio_data = self.generate_variation(midi_id)
|
| 104 |
-
examples[midi_id] = (midi_data, audio_data)
|
| 105 |
-
self.example_variations = examples
|
| 106 |
-
return examples
|
| 107 |
|
| 108 |
def load_midi(self, file_path):
|
| 109 |
midi = MIDI.load(file_path)
|
| 110 |
-
midi_id = f"midi_{len(self.loaded_midi)
|
| 111 |
self.loaded_midi[midi_id] = (file_path, midi)
|
| 112 |
return midi_id
|
| 113 |
|
|
@@ -122,13 +77,13 @@ class MIDIManager:
|
|
| 122 |
instruments.add(event.program)
|
| 123 |
return notes, list(instruments)
|
| 124 |
|
| 125 |
-
def generate_variation(self, midi_id, length_factor=
|
| 126 |
if midi_id not in self.loaded_midi:
|
| 127 |
-
return None
|
| 128 |
_, midi = self.loaded_midi[midi_id]
|
| 129 |
notes, instruments = self.extract_notes_and_instruments(midi)
|
| 130 |
new_notes = []
|
| 131 |
-
for _ in range(int(length_factor)):
|
| 132 |
for note, vel, time in notes:
|
| 133 |
if random.random() < variation:
|
| 134 |
new_note = min(127, max(0, note + random.randint(-2, 2)))
|
|
@@ -147,75 +102,91 @@ class MIDIManager:
|
|
| 147 |
midi_output = io.BytesIO()
|
| 148 |
new_midi.writeFile(midi_output)
|
| 149 |
midi_data = base64.b64encode(midi_output.getvalue()).decode('utf-8')
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
with open(temp_midi, 'wb') as f:
|
| 153 |
-
f.write(midi_output.getvalue())
|
| 154 |
-
audio_output = io.BytesIO()
|
| 155 |
-
self.synthesizer.play_midi(new_midi)
|
| 156 |
-
audio_data = None # Placeholder; see Notes
|
| 157 |
-
if os.path.exists(temp_midi):
|
| 158 |
-
os.remove(temp_midi)
|
| 159 |
-
|
| 160 |
-
self.modified_files.append((midi_data, audio_data))
|
| 161 |
-
return midi_data, audio_data
|
| 162 |
|
| 163 |
-
def
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 171 |
midi_output = io.BytesIO()
|
| 172 |
-
|
| 173 |
midi_data = base64.b64encode(midi_output.getvalue()).decode('utf-8')
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
with open(temp_midi, 'wb') as f:
|
| 177 |
-
f.write(midi_output.getvalue())
|
| 178 |
-
audio_output = io.BytesIO()
|
| 179 |
-
self.synthesizer.play_midi(midi)
|
| 180 |
-
audio_data = None # Placeholder
|
| 181 |
-
if os.path.exists(temp_midi):
|
| 182 |
-
os.remove(temp_midi)
|
| 183 |
-
|
| 184 |
-
self.modified_files.append((midi_data, audio_data))
|
| 185 |
-
return midi_data, audio_data
|
| 186 |
|
| 187 |
def play_with_loop(self, midi_data):
|
| 188 |
self.is_playing = True
|
| 189 |
midi_file = MIDI.load(io.BytesIO(base64.b64decode(midi_data)))
|
| 190 |
while self.is_playing:
|
| 191 |
self.synthesizer.play_midi(midi_file)
|
| 192 |
-
return "Stopped"
|
| 193 |
|
| 194 |
def stop_playback(self):
|
| 195 |
self.is_playing = False
|
| 196 |
-
return "
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 197 |
|
| 198 |
def create_download_list():
|
| 199 |
html = "<h3>Downloads</h3><ul>"
|
| 200 |
-
for i,
|
| 201 |
-
html += f'<li><a href="data:audio/midi;base64,{midi_data}" download="midi_{i}.mid">MIDI {i}</a>'
|
| 202 |
-
if audio_data:
|
| 203 |
-
html += f' | <a href="data:audio/wav;base64,{audio_data}" download="audio_{i}.wav">Audio {i}</a>'
|
| 204 |
-
html += '</li>'
|
| 205 |
html += "</ul>"
|
| 206 |
return html
|
| 207 |
|
| 208 |
-
def get_midi_choices():
|
| 209 |
-
return [(os.path.basename(path), midi_id) for midi_id, (path, _) in midi_processor.loaded_midi.items()]
|
| 210 |
-
|
| 211 |
-
def get_example_choices():
|
| 212 |
-
return [(os.path.basename(path), midi_id) for midi_id, (path, _) in midi_processor.loaded_midi.items() if midi_id.startswith("example")]
|
| 213 |
-
|
| 214 |
if __name__ == "__main__":
|
| 215 |
parser = argparse.ArgumentParser()
|
| 216 |
parser.add_argument("--port", type=int, default=7860)
|
| 217 |
parser.add_argument("--share", action="store_true")
|
| 218 |
-
parser.add_argument("--batch", type=int, default=1)
|
| 219 |
opt = parser.parse_args()
|
| 220 |
|
| 221 |
midi_manager = MIDIDeviceManager()
|
|
@@ -227,96 +198,50 @@ if __name__ == "__main__":
|
|
| 227 |
with gr.Tabs():
|
| 228 |
# Tab 1: MIDI Prompt (Main Tab)
|
| 229 |
with gr.Tab("MIDI Prompt"):
|
| 230 |
-
|
| 231 |
-
|
| 232 |
-
|
| 233 |
|
| 234 |
-
def
|
| 235 |
-
|
|
|
|
| 236 |
midi_data = None
|
| 237 |
-
for file in files
|
| 238 |
midi_id = midi_processor.load_midi(file.name)
|
| 239 |
-
|
| 240 |
-
midi_data
|
| 241 |
-
|
|
|
|
| 242 |
|
| 243 |
-
|
| 244 |
-
|
| 245 |
|
| 246 |
-
# Tab 2:
|
| 247 |
-
with gr.Tab("
|
| 248 |
-
|
| 249 |
-
example_output = gr.Audio(label="Example Preview", type="bytes", autoplay=True)
|
| 250 |
-
|
| 251 |
-
def load_example(midi_id):
|
| 252 |
-
if not midi_id or midi_id not in midi_processor.example_variations:
|
| 253 |
-
return None
|
| 254 |
-
midi_data, audio_data = midi_processor.example_variations[midi_id]
|
| 255 |
-
midi_processor.play_with_loop(midi_data)
|
| 256 |
-
return io.BytesIO(base64.b64decode(midi_data))
|
| 257 |
-
|
| 258 |
-
example_select.change(load_example, inputs=[example_select], outputs=[example_output])
|
| 259 |
|
| 260 |
-
# Tab 3:
|
| 261 |
-
with gr.Tab("
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
variation = gr.Slider(0, 1, value=0.3, label="Variation")
|
| 265 |
-
generate_btn = gr.Button("Generate")
|
| 266 |
-
effect = gr.Radio(["tempo"], label="Synth Effect", value="tempo")
|
| 267 |
-
intensity = gr.Slider(0, 1, value=0.5, label="Effect Intensity")
|
| 268 |
-
apply_btn = gr.Button("Apply Effect")
|
| 269 |
stop_btn = gr.Button("Stop Playback")
|
| 270 |
-
output = gr.Audio(label="Preview", type="bytes", autoplay=True)
|
| 271 |
-
status = gr.Textbox(label="Status", value="Ready")
|
| 272 |
-
midi_device = gr.Dropdown(label="MIDI Output Device", choices=midi_manager.get_output_devices(), type="index")
|
| 273 |
-
tempo = gr.Slider(label="Tempo (BPM)", minimum=40, maximum=200, value=120, step=1)
|
| 274 |
-
device_info = gr.Textbox(label="Connected MIDI Devices", value=midi_manager.get_device_info(), readonly=True)
|
| 275 |
-
refresh_btn = gr.Button("🔄 Refresh MIDI Devices")
|
| 276 |
-
|
| 277 |
-
def update_dropdown(choices):
|
| 278 |
-
return gr.update(choices=choices)
|
| 279 |
-
|
| 280 |
-
gr.State(get_midi_choices()).change(update_dropdown, inputs=[gr.State()], outputs=[midi_select])
|
| 281 |
-
|
| 282 |
-
def generate(midi_id, length, var):
|
| 283 |
-
if not midi_id:
|
| 284 |
-
return None, "Select a MIDI file", create_download_list()
|
| 285 |
-
midi_data, audio_data = midi_processor.generate_variation(midi_id, length, var)
|
| 286 |
-
midi_processor.play_with_loop(midi_data)
|
| 287 |
-
return io.BytesIO(base64.b64decode(midi_data)), "Playing", create_download_list()
|
| 288 |
-
|
| 289 |
-
def apply_effect(midi_data, fx, inten):
|
| 290 |
-
if not midi_data:
|
| 291 |
-
return None, "Generate a MIDI first", create_download_list()
|
| 292 |
-
new_midi_data, audio_data = midi_processor.apply_synth_effect(midi_data.decode('utf-8'), fx, inten)
|
| 293 |
-
midi_processor.play_with_loop(new_midi_data)
|
| 294 |
-
return io.BytesIO(base64.b64decode(new_midi_data)), "Playing", create_download_list()
|
| 295 |
|
| 296 |
def refresh_devices():
|
| 297 |
-
return midi_manager.
|
| 298 |
|
| 299 |
-
|
| 300 |
-
outputs=[output, status, "downloads"])
|
| 301 |
-
apply_btn.click(apply_effect, inputs=[output, effect, intensity],
|
| 302 |
-
outputs=[output, status, "downloads"])
|
| 303 |
stop_btn.click(midi_processor.stop_playback, inputs=None, outputs=[status])
|
| 304 |
-
refresh_btn.click(refresh_devices, inputs=None, outputs=[midi_device, device_info])
|
| 305 |
-
|
| 306 |
-
# Tab 4: Downloads
|
| 307 |
-
with gr.Tab("Downloads", elem_id="downloads"):
|
| 308 |
-
downloads = gr.HTML(value=create_download_list())
|
| 309 |
|
| 310 |
gr.Markdown("""
|
| 311 |
<div style='text-align: center; margin-top: 20px;'>
|
| 312 |
<img src='https://huggingface.co/front/assets/huggingface_logo-noborder.svg' alt='Hugging Face Logo' style='width: 50px;'><br>
|
| 313 |
<strong>Hugging Face</strong><br>
|
| 314 |
-
<a href='https://huggingface.co/models'>Models</a> |
|
| 315 |
-
<a href='https://huggingface.co/datasets'>Datasets</a> |
|
| 316 |
-
<a href='https://huggingface.co/spaces'>Spaces</a> |
|
| 317 |
-
<a href='https://huggingface.co/posts'>Posts</a> |
|
| 318 |
-
<a href='https://huggingface.co/docs'>Docs</a> |
|
| 319 |
-
<a href='https://huggingface.co/enterprise'>Enterprise</a> |
|
| 320 |
<a href='https://huggingface.co/pricing'>Pricing</a>
|
| 321 |
</div>
|
| 322 |
""")
|
|
|
|
| 2 |
import argparse
|
| 3 |
import os
|
| 4 |
import glob
|
| 5 |
+
import json
|
| 6 |
import rtmidi
|
| 7 |
import gradio as gr
|
| 8 |
import numpy as np
|
| 9 |
+
import onnxruntime as rt
|
|
|
|
|
|
|
|
|
|
| 10 |
from huggingface_hub import hf_hub_download
|
| 11 |
+
import MIDI
|
| 12 |
from midi_synthesizer import MidiSynthesizer
|
| 13 |
+
from midi_tokenizer import MIDITokenizer
|
| 14 |
|
| 15 |
MAX_SEED = np.iinfo(np.int32).max
|
| 16 |
in_space = os.getenv("SYSTEM") == "spaces"
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
class MIDIDeviceManager:
|
| 19 |
def __init__(self):
|
| 20 |
self.midiout = rtmidi.MidiOut()
|
|
|
|
| 45 |
def __init__(self):
|
| 46 |
self.soundfont_path = hf_hub_download(repo_id="skytnt/midi-model", filename="soundfont.sf2")
|
| 47 |
self.synthesizer = MidiSynthesizer(self.soundfont_path)
|
| 48 |
+
self.loaded_midi = {}
|
| 49 |
+
self.modified_files = []
|
|
|
|
| 50 |
self.is_playing = False
|
| 51 |
+
self.tokenizer = self.load_tokenizer("skytnt/midi-model")
|
| 52 |
+
self.model_base = rt.InferenceSession(hf_hub_download(repo_id="skytnt/midi-model", filename="onnx/model_base.onnx"), providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
|
| 53 |
+
self.model_token = rt.InferenceSession(hf_hub_download(repo_id="skytnt/midi-model", filename="onnx/model_token.onnx"), providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
|
| 54 |
+
|
| 55 |
+
def load_tokenizer(self, repo_id):
|
| 56 |
+
config_path = hf_hub_download(repo_id=repo_id, filename="config.json")
|
| 57 |
+
with open(config_path, "r") as f:
|
| 58 |
+
config = json.load(f)
|
| 59 |
+
tokenizer = MIDITokenizer(config["tokenizer"]["version"])
|
| 60 |
+
tokenizer.set_optimise_midi(config["tokenizer"]["optimise_midi"])
|
| 61 |
+
return tokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
def load_midi(self, file_path):
|
| 64 |
midi = MIDI.load(file_path)
|
| 65 |
+
midi_id = f"midi_{len(self.loaded_midi)}"
|
| 66 |
self.loaded_midi[midi_id] = (file_path, midi)
|
| 67 |
return midi_id
|
| 68 |
|
|
|
|
| 77 |
instruments.add(event.program)
|
| 78 |
return notes, list(instruments)
|
| 79 |
|
| 80 |
+
def generate_variation(self, midi_id, length_factor=10, variation=0.3):
|
| 81 |
if midi_id not in self.loaded_midi:
|
| 82 |
+
return None
|
| 83 |
_, midi = self.loaded_midi[midi_id]
|
| 84 |
notes, instruments = self.extract_notes_and_instruments(midi)
|
| 85 |
new_notes = []
|
| 86 |
+
for _ in range(int(length_factor)): # Max length: 10x repetition
|
| 87 |
for note, vel, time in notes:
|
| 88 |
if random.random() < variation:
|
| 89 |
new_note = min(127, max(0, note + random.randint(-2, 2)))
|
|
|
|
| 102 |
midi_output = io.BytesIO()
|
| 103 |
new_midi.writeFile(midi_output)
|
| 104 |
midi_data = base64.b64encode(midi_output.getvalue()).decode('utf-8')
|
| 105 |
+
self.modified_files.append(midi_data)
|
| 106 |
+
return midi_data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
|
| 108 |
+
def generate_onnx(self, midi_id, max_len=1024, temp=1.0, top_p=0.98, top_k=20):
|
| 109 |
+
if midi_id not in self.loaded_midi:
|
| 110 |
+
return None
|
| 111 |
+
_, mid = self.loaded_midi[midi_id]
|
| 112 |
+
mid_seq = self.tokenizer.tokenize(MIDI.midi2score(mid))
|
| 113 |
+
mid = np.asarray([mid_seq], dtype=np.int64)
|
| 114 |
+
generator = np.random.RandomState(random.randint(0, MAX_SEED))
|
| 115 |
|
| 116 |
+
# Simplified ONNX generation from app_onnx.py
|
| 117 |
+
input_tensor = mid
|
| 118 |
+
cur_len = input_tensor.shape[1]
|
| 119 |
+
model = [self.model_base, self.model_token, self.tokenizer]
|
| 120 |
+
|
| 121 |
+
while cur_len < max_len:
|
| 122 |
+
inputs = {"x": rt.OrtValue.ortvalue_from_numpy(input_tensor[:, -1:], device_type="cuda")}
|
| 123 |
+
outputs = {"hidden": rt.OrtValue.ortvalue_from_shape_and_type((1, 1, 1024), np.float32, device_type="cuda")}
|
| 124 |
+
io_binding = model[0].io_binding()
|
| 125 |
+
for name, val in inputs.items():
|
| 126 |
+
io_binding.bind_ortvalue_input(name, val)
|
| 127 |
+
for name in outputs:
|
| 128 |
+
io_binding.bind_ortvalue_output(name, outputs[name])
|
| 129 |
+
model[0].run_with_iobinding(io_binding)
|
| 130 |
+
hidden = outputs["hidden"].numpy()[:, -1:]
|
| 131 |
+
|
| 132 |
+
logits = model[1].run(None, {"hidden": hidden})[0]
|
| 133 |
+
scores = softmax(logits / temp, -1)
|
| 134 |
+
next_token = sample_top_p_k(scores, top_p, top_k, generator)
|
| 135 |
+
input_tensor = np.concatenate([input_tensor, next_token], axis=1)
|
| 136 |
+
cur_len += 1
|
| 137 |
+
|
| 138 |
+
mid_seq = input_tensor.tolist()[0]
|
| 139 |
+
new_midi = self.tokenizer.detokenize(mid_seq)
|
| 140 |
midi_output = io.BytesIO()
|
| 141 |
+
MIDI.score2midi(new_midi, midi_output)
|
| 142 |
midi_data = base64.b64encode(midi_output.getvalue()).decode('utf-8')
|
| 143 |
+
self.modified_files.append(midi_data)
|
| 144 |
+
return midi_data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
|
| 146 |
def play_with_loop(self, midi_data):
|
| 147 |
self.is_playing = True
|
| 148 |
midi_file = MIDI.load(io.BytesIO(base64.b64decode(midi_data)))
|
| 149 |
while self.is_playing:
|
| 150 |
self.synthesizer.play_midi(midi_file)
|
|
|
|
| 151 |
|
| 152 |
def stop_playback(self):
|
| 153 |
self.is_playing = False
|
| 154 |
+
return "Playback stopped"
|
| 155 |
+
|
| 156 |
+
def softmax(x, axis):
|
| 157 |
+
x_max = np.max(x, axis=axis, keepdims=True)
|
| 158 |
+
exp_x_shifted = np.exp(x - x_max)
|
| 159 |
+
return exp_x_shifted / np.sum(exp_x_shifted, axis=axis, keepdims=True)
|
| 160 |
+
|
| 161 |
+
def sample_top_p_k(probs, p, k, generator=None):
|
| 162 |
+
if generator is None:
|
| 163 |
+
generator = np.random
|
| 164 |
+
probs_idx = np.argsort(-probs, axis=-1)
|
| 165 |
+
probs_sort = np.take_along_axis(probs, probs_idx, -1)
|
| 166 |
+
probs_sum = np.cumsum(probs_sort, axis=-1)
|
| 167 |
+
mask = probs_sum - probs_sort > p
|
| 168 |
+
probs_sort[mask] = 0.0
|
| 169 |
+
mask = np.zeros(probs_sort.shape[-1])
|
| 170 |
+
mask[:k] = 1
|
| 171 |
+
probs_sort *= mask
|
| 172 |
+
probs_sort /= np.sum(probs_sort, axis=-1, keepdims=True)
|
| 173 |
+
shape = probs_sort.shape
|
| 174 |
+
probs_sort_flat = probs_sort.reshape(-1, shape[-1])
|
| 175 |
+
probs_idx_flat = probs_idx.reshape(-1, shape[-1])
|
| 176 |
+
next_token = np.stack([generator.choice(idxs, p=pvals) for pvals, idxs in zip(probs_sort_flat, probs_idx_flat)])
|
| 177 |
+
return next_token.reshape(*shape[:-1])
|
| 178 |
|
| 179 |
def create_download_list():
|
| 180 |
html = "<h3>Downloads</h3><ul>"
|
| 181 |
+
for i, midi_data in enumerate(midi_processor.modified_files):
|
| 182 |
+
html += f'<li><a href="data:audio/midi;base64,{midi_data}" download="midi_{i}.mid">MIDI {i}</a></li>'
|
|
|
|
|
|
|
|
|
|
| 183 |
html += "</ul>"
|
| 184 |
return html
|
| 185 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 186 |
if __name__ == "__main__":
|
| 187 |
parser = argparse.ArgumentParser()
|
| 188 |
parser.add_argument("--port", type=int, default=7860)
|
| 189 |
parser.add_argument("--share", action="store_true")
|
|
|
|
| 190 |
opt = parser.parse_args()
|
| 191 |
|
| 192 |
midi_manager = MIDIDeviceManager()
|
|
|
|
| 198 |
with gr.Tabs():
|
| 199 |
# Tab 1: MIDI Prompt (Main Tab)
|
| 200 |
with gr.Tab("MIDI Prompt"):
|
| 201 |
+
midi_upload = gr.File(label="Upload MIDI File", file_count="multiple")
|
| 202 |
+
output = gr.Audio(label="Generated MIDI", type="bytes", autoplay=True)
|
| 203 |
+
status = gr.Textbox(label="Status", value="Ready", interactive=False)
|
| 204 |
|
| 205 |
+
def process_midi(files):
|
| 206 |
+
if not files:
|
| 207 |
+
return None, "No file uploaded"
|
| 208 |
midi_data = None
|
| 209 |
+
for file in files:
|
| 210 |
midi_id = midi_processor.load_midi(file.name)
|
| 211 |
+
# Use ONNX generation for advanced synthesis
|
| 212 |
+
midi_data = midi_processor.generate_onnx(midi_id, max_len=1024)
|
| 213 |
+
midi_processor.play_with_loop(midi_data)
|
| 214 |
+
return io.BytesIO(base64.b64decode(midi_data)), "Playing", create_download_list()
|
| 215 |
|
| 216 |
+
midi_upload.change(process_midi, inputs=[midi_upload],
|
| 217 |
+
outputs=[output, status, "downloads"])
|
| 218 |
|
| 219 |
+
# Tab 2: Downloads
|
| 220 |
+
with gr.Tab("Downloads", elem_id="downloads"):
|
| 221 |
+
downloads = gr.HTML(value="No generated files yet")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 222 |
|
| 223 |
+
# Tab 3: Devices
|
| 224 |
+
with gr.Tab("Devices"):
|
| 225 |
+
device_info = gr.Textbox(label="Connected MIDI Devices", value=midi_manager.get_device_info(), interactive=False)
|
| 226 |
+
refresh_btn = gr.Button("Refresh Devices")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
stop_btn = gr.Button("Stop Playback")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 228 |
|
| 229 |
def refresh_devices():
|
| 230 |
+
return midi_manager.get_device_info()
|
| 231 |
|
| 232 |
+
refresh_btn.click(refresh_devices, inputs=None, outputs=[device_info])
|
|
|
|
|
|
|
|
|
|
| 233 |
stop_btn.click(midi_processor.stop_playback, inputs=None, outputs=[status])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 234 |
|
| 235 |
gr.Markdown("""
|
| 236 |
<div style='text-align: center; margin-top: 20px;'>
|
| 237 |
<img src='https://huggingface.co/front/assets/huggingface_logo-noborder.svg' alt='Hugging Face Logo' style='width: 50px;'><br>
|
| 238 |
<strong>Hugging Face</strong><br>
|
| 239 |
+
<a href='https://huggingface.co/models'>Models</a> |
|
| 240 |
+
<a href='https://huggingface.co/datasets'>Datasets</a> |
|
| 241 |
+
<a href='https://huggingface.co/spaces'>Spaces</a> |
|
| 242 |
+
<a href='https://huggingface.co/posts'>Posts</a> |
|
| 243 |
+
<a href='https://huggingface.co/docs'>Docs</a> |
|
| 244 |
+
<a href='https://huggingface.co/enterprise'>Enterprise</a> |
|
| 245 |
<a href='https://huggingface.co/pricing'>Pricing</a>
|
| 246 |
</div>
|
| 247 |
""")
|