Simon-Kotchou commited on
Commit
000fdb2
1 Parent(s): c583bc6

Upload ASTForAudioClassification

Browse files
Files changed (2) hide show
  1. config.json +1081 -23
  2. model.safetensors +2 -2
config.json CHANGED
@@ -1,23 +1,1081 @@
1
- {
2
- "architectures": [
3
- "ASTForAudioClassification"
4
- ],
5
- "attention_probs_dropout_prob": 0.0,
6
- "frequency_stride": 16,
7
- "hidden_act": "gelu",
8
- "hidden_dropout_prob": 0.0,
9
- "hidden_size": 192,
10
- "initializer_range": 0.02,
11
- "intermediate_size": 768,
12
- "layer_norm_eps": 1e-12,
13
- "max_length": 1024,
14
- "model_type": "audio-spectrogram-transformer",
15
- "num_attention_heads": 3,
16
- "num_hidden_layers": 12,
17
- "num_mel_bins": 128,
18
- "patch_size": 16,
19
- "qkv_bias": true,
20
- "time_stride": 16,
21
- "torch_dtype": "float32",
22
- "transformers_version": "4.35.0"
23
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "ASTForAudioClassification"
4
+ ],
5
+ "attention_probs_dropout_prob": 0.0,
6
+ "frequency_stride": 16,
7
+ "hidden_act": "gelu",
8
+ "hidden_dropout_prob": 0.0,
9
+ "hidden_size": 192,
10
+ "id2label": {
11
+ "0": "Speech",
12
+ "1": "Male speech, man speaking",
13
+ "2": "Female speech, woman speaking",
14
+ "3": "Child speech, kid speaking",
15
+ "4": "Conversation",
16
+ "5": "Narration, monologue",
17
+ "6": "Babbling",
18
+ "7": "Speech synthesizer",
19
+ "8": "Shout",
20
+ "9": "Bellow",
21
+ "10": "Whoop",
22
+ "11": "Yell",
23
+ "12": "Battle cry",
24
+ "13": "Children shouting",
25
+ "14": "Screaming",
26
+ "15": "Whispering",
27
+ "16": "Laughter",
28
+ "17": "Baby laughter",
29
+ "18": "Giggle",
30
+ "19": "Snicker",
31
+ "20": "Belly laugh",
32
+ "21": "Chuckle, chortle",
33
+ "22": "Crying, sobbing",
34
+ "23": "Baby cry, infant cry",
35
+ "24": "Whimper",
36
+ "25": "Wail, moan",
37
+ "26": "Sigh",
38
+ "27": "Singing",
39
+ "28": "Choir",
40
+ "29": "Yodeling",
41
+ "30": "Chant",
42
+ "31": "Mantra",
43
+ "32": "Male singing",
44
+ "33": "Female singing",
45
+ "34": "Child singing",
46
+ "35": "Synthetic singing",
47
+ "36": "Rapping",
48
+ "37": "Humming",
49
+ "38": "Groan",
50
+ "39": "Grunt",
51
+ "40": "Whistling",
52
+ "41": "Breathing",
53
+ "42": "Wheeze",
54
+ "43": "Snoring",
55
+ "44": "Gasp",
56
+ "45": "Pant",
57
+ "46": "Snort",
58
+ "47": "Cough",
59
+ "48": "Throat clearing",
60
+ "49": "Sneeze",
61
+ "50": "Sniff",
62
+ "51": "Run",
63
+ "52": "Shuffle",
64
+ "53": "Walk, footsteps",
65
+ "54": "Chewing, mastication",
66
+ "55": "Biting",
67
+ "56": "Gargling",
68
+ "57": "Stomach rumble",
69
+ "58": "Burping, eructation",
70
+ "59": "Hiccup",
71
+ "60": "Fart",
72
+ "61": "Hands",
73
+ "62": "Finger snapping",
74
+ "63": "Clapping",
75
+ "64": "Heart sounds, heartbeat",
76
+ "65": "Heart murmur",
77
+ "66": "Cheering",
78
+ "67": "Applause",
79
+ "68": "Chatter",
80
+ "69": "Crowd",
81
+ "70": "Hubbub, speech noise, speech babble",
82
+ "71": "Children playing",
83
+ "72": "Animal",
84
+ "73": "Domestic animals, pets",
85
+ "74": "Dog",
86
+ "75": "Bark",
87
+ "76": "Yip",
88
+ "77": "Howl",
89
+ "78": "Bow-wow",
90
+ "79": "Growling",
91
+ "80": "Whimper (dog)",
92
+ "81": "Cat",
93
+ "82": "Purr",
94
+ "83": "Meow",
95
+ "84": "Hiss",
96
+ "85": "Caterwaul",
97
+ "86": "Livestock, farm animals, working animals",
98
+ "87": "Horse",
99
+ "88": "Clip-clop",
100
+ "89": "Neigh, whinny",
101
+ "90": "Cattle, bovinae",
102
+ "91": "Moo",
103
+ "92": "Cowbell",
104
+ "93": "Pig",
105
+ "94": "Oink",
106
+ "95": "Goat",
107
+ "96": "Bleat",
108
+ "97": "Sheep",
109
+ "98": "Fowl",
110
+ "99": "Chicken, rooster",
111
+ "100": "Cluck",
112
+ "101": "Crowing, cock-a-doodle-doo",
113
+ "102": "Turkey",
114
+ "103": "Gobble",
115
+ "104": "Duck",
116
+ "105": "Quack",
117
+ "106": "Goose",
118
+ "107": "Honk",
119
+ "108": "Wild animals",
120
+ "109": "Roaring cats (lions, tigers)",
121
+ "110": "Roar",
122
+ "111": "Bird",
123
+ "112": "Bird vocalization, bird call, bird song",
124
+ "113": "Chirp, tweet",
125
+ "114": "Squawk",
126
+ "115": "Pigeon, dove",
127
+ "116": "Coo",
128
+ "117": "Crow",
129
+ "118": "Caw",
130
+ "119": "Owl",
131
+ "120": "Hoot",
132
+ "121": "Bird flight, flapping wings",
133
+ "122": "Canidae, dogs, wolves",
134
+ "123": "Rodents, rats, mice",
135
+ "124": "Mouse",
136
+ "125": "Patter",
137
+ "126": "Insect",
138
+ "127": "Cricket",
139
+ "128": "Mosquito",
140
+ "129": "Fly, housefly",
141
+ "130": "Buzz",
142
+ "131": "Bee, wasp, etc.",
143
+ "132": "Frog",
144
+ "133": "Croak",
145
+ "134": "Snake",
146
+ "135": "Rattle",
147
+ "136": "Whale vocalization",
148
+ "137": "Music",
149
+ "138": "Musical instrument",
150
+ "139": "Plucked string instrument",
151
+ "140": "Guitar",
152
+ "141": "Electric guitar",
153
+ "142": "Bass guitar",
154
+ "143": "Acoustic guitar",
155
+ "144": "Steel guitar, slide guitar",
156
+ "145": "Tapping (guitar technique)",
157
+ "146": "Strum",
158
+ "147": "Banjo",
159
+ "148": "Sitar",
160
+ "149": "Mandolin",
161
+ "150": "Zither",
162
+ "151": "Ukulele",
163
+ "152": "Keyboard (musical)",
164
+ "153": "Piano",
165
+ "154": "Electric piano",
166
+ "155": "Organ",
167
+ "156": "Electronic organ",
168
+ "157": "Hammond organ",
169
+ "158": "Synthesizer",
170
+ "159": "Sampler",
171
+ "160": "Harpsichord",
172
+ "161": "Percussion",
173
+ "162": "Drum kit",
174
+ "163": "Drum machine",
175
+ "164": "Drum",
176
+ "165": "Snare drum",
177
+ "166": "Rimshot",
178
+ "167": "Drum roll",
179
+ "168": "Bass drum",
180
+ "169": "Timpani",
181
+ "170": "Tabla",
182
+ "171": "Cymbal",
183
+ "172": "Hi-hat",
184
+ "173": "Wood block",
185
+ "174": "Tambourine",
186
+ "175": "Rattle (instrument)",
187
+ "176": "Maraca",
188
+ "177": "Gong",
189
+ "178": "Tubular bells",
190
+ "179": "Mallet percussion",
191
+ "180": "Marimba, xylophone",
192
+ "181": "Glockenspiel",
193
+ "182": "Vibraphone",
194
+ "183": "Steelpan",
195
+ "184": "Orchestra",
196
+ "185": "Brass instrument",
197
+ "186": "French horn",
198
+ "187": "Trumpet",
199
+ "188": "Trombone",
200
+ "189": "Bowed string instrument",
201
+ "190": "String section",
202
+ "191": "Violin, fiddle",
203
+ "192": "Pizzicato",
204
+ "193": "Cello",
205
+ "194": "Double bass",
206
+ "195": "Wind instrument, woodwind instrument",
207
+ "196": "Flute",
208
+ "197": "Saxophone",
209
+ "198": "Clarinet",
210
+ "199": "Harp",
211
+ "200": "Bell",
212
+ "201": "Church bell",
213
+ "202": "Jingle bell",
214
+ "203": "Bicycle bell",
215
+ "204": "Tuning fork",
216
+ "205": "Chime",
217
+ "206": "Wind chime",
218
+ "207": "Change ringing (campanology)",
219
+ "208": "Harmonica",
220
+ "209": "Accordion",
221
+ "210": "Bagpipes",
222
+ "211": "Didgeridoo",
223
+ "212": "Shofar",
224
+ "213": "Theremin",
225
+ "214": "Singing bowl",
226
+ "215": "Scratching (performance technique)",
227
+ "216": "Pop music",
228
+ "217": "Hip hop music",
229
+ "218": "Beatboxing",
230
+ "219": "Rock music",
231
+ "220": "Heavy metal",
232
+ "221": "Punk rock",
233
+ "222": "Grunge",
234
+ "223": "Progressive rock",
235
+ "224": "Rock and roll",
236
+ "225": "Psychedelic rock",
237
+ "226": "Rhythm and blues",
238
+ "227": "Soul music",
239
+ "228": "Reggae",
240
+ "229": "Country",
241
+ "230": "Swing music",
242
+ "231": "Bluegrass",
243
+ "232": "Funk",
244
+ "233": "Folk music",
245
+ "234": "Middle Eastern music",
246
+ "235": "Jazz",
247
+ "236": "Disco",
248
+ "237": "Classical music",
249
+ "238": "Opera",
250
+ "239": "Electronic music",
251
+ "240": "House music",
252
+ "241": "Techno",
253
+ "242": "Dubstep",
254
+ "243": "Drum and bass",
255
+ "244": "Electronica",
256
+ "245": "Electronic dance music",
257
+ "246": "Ambient music",
258
+ "247": "Trance music",
259
+ "248": "Music of Latin America",
260
+ "249": "Salsa music",
261
+ "250": "Flamenco",
262
+ "251": "Blues",
263
+ "252": "Music for children",
264
+ "253": "New-age music",
265
+ "254": "Vocal music",
266
+ "255": "A capella",
267
+ "256": "Music of Africa",
268
+ "257": "Afrobeat",
269
+ "258": "Christian music",
270
+ "259": "Gospel music",
271
+ "260": "Music of Asia",
272
+ "261": "Carnatic music",
273
+ "262": "Music of Bollywood",
274
+ "263": "Ska",
275
+ "264": "Traditional music",
276
+ "265": "Independent music",
277
+ "266": "Song",
278
+ "267": "Background music",
279
+ "268": "Theme music",
280
+ "269": "Jingle (music)",
281
+ "270": "Soundtrack music",
282
+ "271": "Lullaby",
283
+ "272": "Video game music",
284
+ "273": "Christmas music",
285
+ "274": "Dance music",
286
+ "275": "Wedding music",
287
+ "276": "Happy music",
288
+ "277": "Funny music",
289
+ "278": "Sad music",
290
+ "279": "Tender music",
291
+ "280": "Exciting music",
292
+ "281": "Angry music",
293
+ "282": "Scary music",
294
+ "283": "Wind",
295
+ "284": "Rustling leaves",
296
+ "285": "Wind noise (microphone)",
297
+ "286": "Thunderstorm",
298
+ "287": "Thunder",
299
+ "288": "Water",
300
+ "289": "Rain",
301
+ "290": "Raindrop",
302
+ "291": "Rain on surface",
303
+ "292": "Stream",
304
+ "293": "Waterfall",
305
+ "294": "Ocean",
306
+ "295": "Waves, surf",
307
+ "296": "Steam",
308
+ "297": "Gurgling",
309
+ "298": "Fire",
310
+ "299": "Crackle",
311
+ "300": "Vehicle",
312
+ "301": "Boat, Water vehicle",
313
+ "302": "Sailboat, sailing ship",
314
+ "303": "Rowboat, canoe, kayak",
315
+ "304": "Motorboat, speedboat",
316
+ "305": "Ship",
317
+ "306": "Motor vehicle (road)",
318
+ "307": "Car",
319
+ "308": "Vehicle horn, car horn, honking",
320
+ "309": "Toot",
321
+ "310": "Car alarm",
322
+ "311": "Power windows, electric windows",
323
+ "312": "Skidding",
324
+ "313": "Tire squeal",
325
+ "314": "Car passing by",
326
+ "315": "Race car, auto racing",
327
+ "316": "Truck",
328
+ "317": "Air brake",
329
+ "318": "Air horn, truck horn",
330
+ "319": "Reversing beeps",
331
+ "320": "Ice cream truck, ice cream van",
332
+ "321": "Bus",
333
+ "322": "Emergency vehicle",
334
+ "323": "Police car (siren)",
335
+ "324": "Ambulance (siren)",
336
+ "325": "Fire engine, fire truck (siren)",
337
+ "326": "Motorcycle",
338
+ "327": "Traffic noise, roadway noise",
339
+ "328": "Rail transport",
340
+ "329": "Train",
341
+ "330": "Train whistle",
342
+ "331": "Train horn",
343
+ "332": "Railroad car, train wagon",
344
+ "333": "Train wheels squealing",
345
+ "334": "Subway, metro, underground",
346
+ "335": "Aircraft",
347
+ "336": "Aircraft engine",
348
+ "337": "Jet engine",
349
+ "338": "Propeller, airscrew",
350
+ "339": "Helicopter",
351
+ "340": "Fixed-wing aircraft, airplane",
352
+ "341": "Bicycle",
353
+ "342": "Skateboard",
354
+ "343": "Engine",
355
+ "344": "Light engine (high frequency)",
356
+ "345": "Dental drill, dentist's drill",
357
+ "346": "Lawn mower",
358
+ "347": "Chainsaw",
359
+ "348": "Medium engine (mid frequency)",
360
+ "349": "Heavy engine (low frequency)",
361
+ "350": "Engine knocking",
362
+ "351": "Engine starting",
363
+ "352": "Idling",
364
+ "353": "Accelerating, revving, vroom",
365
+ "354": "Door",
366
+ "355": "Doorbell",
367
+ "356": "Ding-dong",
368
+ "357": "Sliding door",
369
+ "358": "Slam",
370
+ "359": "Knock",
371
+ "360": "Tap",
372
+ "361": "Squeak",
373
+ "362": "Cupboard open or close",
374
+ "363": "Drawer open or close",
375
+ "364": "Dishes, pots, and pans",
376
+ "365": "Cutlery, silverware",
377
+ "366": "Chopping (food)",
378
+ "367": "Frying (food)",
379
+ "368": "Microwave oven",
380
+ "369": "Blender",
381
+ "370": "Water tap, faucet",
382
+ "371": "Sink (filling or washing)",
383
+ "372": "Bathtub (filling or washing)",
384
+ "373": "Hair dryer",
385
+ "374": "Toilet flush",
386
+ "375": "Toothbrush",
387
+ "376": "Electric toothbrush",
388
+ "377": "Vacuum cleaner",
389
+ "378": "Zipper (clothing)",
390
+ "379": "Keys jangling",
391
+ "380": "Coin (dropping)",
392
+ "381": "Scissors",
393
+ "382": "Electric shaver, electric razor",
394
+ "383": "Shuffling cards",
395
+ "384": "Typing",
396
+ "385": "Typewriter",
397
+ "386": "Computer keyboard",
398
+ "387": "Writing",
399
+ "388": "Alarm",
400
+ "389": "Telephone",
401
+ "390": "Telephone bell ringing",
402
+ "391": "Ringtone",
403
+ "392": "Telephone dialing, DTMF",
404
+ "393": "Dial tone",
405
+ "394": "Busy signal",
406
+ "395": "Alarm clock",
407
+ "396": "Siren",
408
+ "397": "Civil defense siren",
409
+ "398": "Buzzer",
410
+ "399": "Smoke detector, smoke alarm",
411
+ "400": "Fire alarm",
412
+ "401": "Foghorn",
413
+ "402": "Whistle",
414
+ "403": "Steam whistle",
415
+ "404": "Mechanisms",
416
+ "405": "Ratchet, pawl",
417
+ "406": "Clock",
418
+ "407": "Tick",
419
+ "408": "Tick-tock",
420
+ "409": "Gears",
421
+ "410": "Pulleys",
422
+ "411": "Sewing machine",
423
+ "412": "Mechanical fan",
424
+ "413": "Air conditioning",
425
+ "414": "Cash register",
426
+ "415": "Printer",
427
+ "416": "Camera",
428
+ "417": "Single-lens reflex camera",
429
+ "418": "Tools",
430
+ "419": "Hammer",
431
+ "420": "Jackhammer",
432
+ "421": "Sawing",
433
+ "422": "Filing (rasp)",
434
+ "423": "Sanding",
435
+ "424": "Power tool",
436
+ "425": "Drill",
437
+ "426": "Explosion",
438
+ "427": "Gunshot, gunfire",
439
+ "428": "Machine gun",
440
+ "429": "Fusillade",
441
+ "430": "Artillery fire",
442
+ "431": "Cap gun",
443
+ "432": "Fireworks",
444
+ "433": "Firecracker",
445
+ "434": "Burst, pop",
446
+ "435": "Eruption",
447
+ "436": "Boom",
448
+ "437": "Wood",
449
+ "438": "Chop",
450
+ "439": "Splinter",
451
+ "440": "Crack",
452
+ "441": "Glass",
453
+ "442": "Chink, clink",
454
+ "443": "Shatter",
455
+ "444": "Liquid",
456
+ "445": "Splash, splatter",
457
+ "446": "Slosh",
458
+ "447": "Squish",
459
+ "448": "Drip",
460
+ "449": "Pour",
461
+ "450": "Trickle, dribble",
462
+ "451": "Gush",
463
+ "452": "Fill (with liquid)",
464
+ "453": "Spray",
465
+ "454": "Pump (liquid)",
466
+ "455": "Stir",
467
+ "456": "Boiling",
468
+ "457": "Sonar",
469
+ "458": "Arrow",
470
+ "459": "Whoosh, swoosh, swish",
471
+ "460": "Thump, thud",
472
+ "461": "Thunk",
473
+ "462": "Electronic tuner",
474
+ "463": "Effects unit",
475
+ "464": "Chorus effect",
476
+ "465": "Basketball bounce",
477
+ "466": "Bang",
478
+ "467": "Slap, smack",
479
+ "468": "Whack, thwack",
480
+ "469": "Smash, crash",
481
+ "470": "Breaking",
482
+ "471": "Bouncing",
483
+ "472": "Whip",
484
+ "473": "Flap",
485
+ "474": "Scratch",
486
+ "475": "Scrape",
487
+ "476": "Rub",
488
+ "477": "Roll",
489
+ "478": "Crushing",
490
+ "479": "Crumpling, crinkling",
491
+ "480": "Tearing",
492
+ "481": "Beep, bleep",
493
+ "482": "Ping",
494
+ "483": "Ding",
495
+ "484": "Clang",
496
+ "485": "Squeal",
497
+ "486": "Creak",
498
+ "487": "Rustle",
499
+ "488": "Whir",
500
+ "489": "Clatter",
501
+ "490": "Sizzle",
502
+ "491": "Clicking",
503
+ "492": "Clickety-clack",
504
+ "493": "Rumble",
505
+ "494": "Plop",
506
+ "495": "Jingle, tinkle",
507
+ "496": "Hum",
508
+ "497": "Zing",
509
+ "498": "Boing",
510
+ "499": "Crunch",
511
+ "500": "Silence",
512
+ "501": "Sine wave",
513
+ "502": "Harmonic",
514
+ "503": "Chirp tone",
515
+ "504": "Sound effect",
516
+ "505": "Pulse",
517
+ "506": "Inside, small room",
518
+ "507": "Inside, large room or hall",
519
+ "508": "Inside, public space",
520
+ "509": "Outside, urban or manmade",
521
+ "510": "Outside, rural or natural",
522
+ "511": "Reverberation",
523
+ "512": "Echo",
524
+ "513": "Noise",
525
+ "514": "Environmental noise",
526
+ "515": "Static",
527
+ "516": "Mains hum",
528
+ "517": "Distortion",
529
+ "518": "Sidetone",
530
+ "519": "Cacophony",
531
+ "520": "White noise",
532
+ "521": "Pink noise",
533
+ "522": "Throbbing",
534
+ "523": "Vibration",
535
+ "524": "Television",
536
+ "525": "Radio",
537
+ "526": "Field recording"
538
+ },
539
+ "initializer_range": 0.02,
540
+ "intermediate_size": 768,
541
+ "label2id": {
542
+ "A capella": 255,
543
+ "Accelerating, revving, vroom": 353,
544
+ "Accordion": 209,
545
+ "Acoustic guitar": 143,
546
+ "Afrobeat": 257,
547
+ "Air brake": 317,
548
+ "Air conditioning": 413,
549
+ "Air horn, truck horn": 318,
550
+ "Aircraft": 335,
551
+ "Aircraft engine": 336,
552
+ "Alarm": 388,
553
+ "Alarm clock": 395,
554
+ "Ambient music": 246,
555
+ "Ambulance (siren)": 324,
556
+ "Angry music": 281,
557
+ "Animal": 72,
558
+ "Applause": 67,
559
+ "Arrow": 458,
560
+ "Artillery fire": 430,
561
+ "Babbling": 6,
562
+ "Baby cry, infant cry": 23,
563
+ "Baby laughter": 17,
564
+ "Background music": 267,
565
+ "Bagpipes": 210,
566
+ "Bang": 466,
567
+ "Banjo": 147,
568
+ "Bark": 75,
569
+ "Basketball bounce": 465,
570
+ "Bass drum": 168,
571
+ "Bass guitar": 142,
572
+ "Bathtub (filling or washing)": 372,
573
+ "Battle cry": 12,
574
+ "Beatboxing": 218,
575
+ "Bee, wasp, etc.": 131,
576
+ "Beep, bleep": 481,
577
+ "Bell": 200,
578
+ "Bellow": 9,
579
+ "Belly laugh": 20,
580
+ "Bicycle": 341,
581
+ "Bicycle bell": 203,
582
+ "Bird": 111,
583
+ "Bird flight, flapping wings": 121,
584
+ "Bird vocalization, bird call, bird song": 112,
585
+ "Biting": 55,
586
+ "Bleat": 96,
587
+ "Blender": 369,
588
+ "Bluegrass": 231,
589
+ "Blues": 251,
590
+ "Boat, Water vehicle": 301,
591
+ "Boiling": 456,
592
+ "Boing": 498,
593
+ "Boom": 436,
594
+ "Bouncing": 471,
595
+ "Bow-wow": 78,
596
+ "Bowed string instrument": 189,
597
+ "Brass instrument": 185,
598
+ "Breaking": 470,
599
+ "Breathing": 41,
600
+ "Burping, eructation": 58,
601
+ "Burst, pop": 434,
602
+ "Bus": 321,
603
+ "Busy signal": 394,
604
+ "Buzz": 130,
605
+ "Buzzer": 398,
606
+ "Cacophony": 519,
607
+ "Camera": 416,
608
+ "Canidae, dogs, wolves": 122,
609
+ "Cap gun": 431,
610
+ "Car": 307,
611
+ "Car alarm": 310,
612
+ "Car passing by": 314,
613
+ "Carnatic music": 261,
614
+ "Cash register": 414,
615
+ "Cat": 81,
616
+ "Caterwaul": 85,
617
+ "Cattle, bovinae": 90,
618
+ "Caw": 118,
619
+ "Cello": 193,
620
+ "Chainsaw": 347,
621
+ "Change ringing (campanology)": 207,
622
+ "Chant": 30,
623
+ "Chatter": 68,
624
+ "Cheering": 66,
625
+ "Chewing, mastication": 54,
626
+ "Chicken, rooster": 99,
627
+ "Child singing": 34,
628
+ "Child speech, kid speaking": 3,
629
+ "Children playing": 71,
630
+ "Children shouting": 13,
631
+ "Chime": 205,
632
+ "Chink, clink": 442,
633
+ "Chirp tone": 503,
634
+ "Chirp, tweet": 113,
635
+ "Choir": 28,
636
+ "Chop": 438,
637
+ "Chopping (food)": 366,
638
+ "Chorus effect": 464,
639
+ "Christian music": 258,
640
+ "Christmas music": 273,
641
+ "Chuckle, chortle": 21,
642
+ "Church bell": 201,
643
+ "Civil defense siren": 397,
644
+ "Clang": 484,
645
+ "Clapping": 63,
646
+ "Clarinet": 198,
647
+ "Classical music": 237,
648
+ "Clatter": 489,
649
+ "Clickety-clack": 492,
650
+ "Clicking": 491,
651
+ "Clip-clop": 88,
652
+ "Clock": 406,
653
+ "Cluck": 100,
654
+ "Coin (dropping)": 380,
655
+ "Computer keyboard": 386,
656
+ "Conversation": 4,
657
+ "Coo": 116,
658
+ "Cough": 47,
659
+ "Country": 229,
660
+ "Cowbell": 92,
661
+ "Crack": 440,
662
+ "Crackle": 299,
663
+ "Creak": 486,
664
+ "Cricket": 127,
665
+ "Croak": 133,
666
+ "Crow": 117,
667
+ "Crowd": 69,
668
+ "Crowing, cock-a-doodle-doo": 101,
669
+ "Crumpling, crinkling": 479,
670
+ "Crunch": 499,
671
+ "Crushing": 478,
672
+ "Crying, sobbing": 22,
673
+ "Cupboard open or close": 362,
674
+ "Cutlery, silverware": 365,
675
+ "Cymbal": 171,
676
+ "Dance music": 274,
677
+ "Dental drill, dentist's drill": 345,
678
+ "Dial tone": 393,
679
+ "Didgeridoo": 211,
680
+ "Ding": 483,
681
+ "Ding-dong": 356,
682
+ "Disco": 236,
683
+ "Dishes, pots, and pans": 364,
684
+ "Distortion": 517,
685
+ "Dog": 74,
686
+ "Domestic animals, pets": 73,
687
+ "Door": 354,
688
+ "Doorbell": 355,
689
+ "Double bass": 194,
690
+ "Drawer open or close": 363,
691
+ "Drill": 425,
692
+ "Drip": 448,
693
+ "Drum": 164,
694
+ "Drum and bass": 243,
695
+ "Drum kit": 162,
696
+ "Drum machine": 163,
697
+ "Drum roll": 167,
698
+ "Dubstep": 242,
699
+ "Duck": 104,
700
+ "Echo": 512,
701
+ "Effects unit": 463,
702
+ "Electric guitar": 141,
703
+ "Electric piano": 154,
704
+ "Electric shaver, electric razor": 382,
705
+ "Electric toothbrush": 376,
706
+ "Electronic dance music": 245,
707
+ "Electronic music": 239,
708
+ "Electronic organ": 156,
709
+ "Electronic tuner": 462,
710
+ "Electronica": 244,
711
+ "Emergency vehicle": 322,
712
+ "Engine": 343,
713
+ "Engine knocking": 350,
714
+ "Engine starting": 351,
715
+ "Environmental noise": 514,
716
+ "Eruption": 435,
717
+ "Exciting music": 280,
718
+ "Explosion": 426,
719
+ "Fart": 60,
720
+ "Female singing": 33,
721
+ "Female speech, woman speaking": 2,
722
+ "Field recording": 526,
723
+ "Filing (rasp)": 422,
724
+ "Fill (with liquid)": 452,
725
+ "Finger snapping": 62,
726
+ "Fire": 298,
727
+ "Fire alarm": 400,
728
+ "Fire engine, fire truck (siren)": 325,
729
+ "Firecracker": 433,
730
+ "Fireworks": 432,
731
+ "Fixed-wing aircraft, airplane": 340,
732
+ "Flamenco": 250,
733
+ "Flap": 473,
734
+ "Flute": 196,
735
+ "Fly, housefly": 129,
736
+ "Foghorn": 401,
737
+ "Folk music": 233,
738
+ "Fowl": 98,
739
+ "French horn": 186,
740
+ "Frog": 132,
741
+ "Frying (food)": 367,
742
+ "Funk": 232,
743
+ "Funny music": 277,
744
+ "Fusillade": 429,
745
+ "Gargling": 56,
746
+ "Gasp": 44,
747
+ "Gears": 409,
748
+ "Giggle": 18,
749
+ "Glass": 441,
750
+ "Glockenspiel": 181,
751
+ "Goat": 95,
752
+ "Gobble": 103,
753
+ "Gong": 177,
754
+ "Goose": 106,
755
+ "Gospel music": 259,
756
+ "Groan": 38,
757
+ "Growling": 79,
758
+ "Grunge": 222,
759
+ "Grunt": 39,
760
+ "Guitar": 140,
761
+ "Gunshot, gunfire": 427,
762
+ "Gurgling": 297,
763
+ "Gush": 451,
764
+ "Hair dryer": 373,
765
+ "Hammer": 419,
766
+ "Hammond organ": 157,
767
+ "Hands": 61,
768
+ "Happy music": 276,
769
+ "Harmonic": 502,
770
+ "Harmonica": 208,
771
+ "Harp": 199,
772
+ "Harpsichord": 160,
773
+ "Heart murmur": 65,
774
+ "Heart sounds, heartbeat": 64,
775
+ "Heavy engine (low frequency)": 349,
776
+ "Heavy metal": 220,
777
+ "Helicopter": 339,
778
+ "Hi-hat": 172,
779
+ "Hiccup": 59,
780
+ "Hip hop music": 217,
781
+ "Hiss": 84,
782
+ "Honk": 107,
783
+ "Hoot": 120,
784
+ "Horse": 87,
785
+ "House music": 240,
786
+ "Howl": 77,
787
+ "Hubbub, speech noise, speech babble": 70,
788
+ "Hum": 496,
789
+ "Humming": 37,
790
+ "Ice cream truck, ice cream van": 320,
791
+ "Idling": 352,
792
+ "Independent music": 265,
793
+ "Insect": 126,
794
+ "Inside, large room or hall": 507,
795
+ "Inside, public space": 508,
796
+ "Inside, small room": 506,
797
+ "Jackhammer": 420,
798
+ "Jazz": 235,
799
+ "Jet engine": 337,
800
+ "Jingle (music)": 269,
801
+ "Jingle bell": 202,
802
+ "Jingle, tinkle": 495,
803
+ "Keyboard (musical)": 152,
804
+ "Keys jangling": 379,
805
+ "Knock": 359,
806
+ "Laughter": 16,
807
+ "Lawn mower": 346,
808
+ "Light engine (high frequency)": 344,
809
+ "Liquid": 444,
810
+ "Livestock, farm animals, working animals": 86,
811
+ "Lullaby": 271,
812
+ "Machine gun": 428,
813
+ "Mains hum": 516,
814
+ "Male singing": 32,
815
+ "Male speech, man speaking": 1,
816
+ "Mallet percussion": 179,
817
+ "Mandolin": 149,
818
+ "Mantra": 31,
819
+ "Maraca": 176,
820
+ "Marimba, xylophone": 180,
821
+ "Mechanical fan": 412,
822
+ "Mechanisms": 404,
823
+ "Medium engine (mid frequency)": 348,
824
+ "Meow": 83,
825
+ "Microwave oven": 368,
826
+ "Middle Eastern music": 234,
827
+ "Moo": 91,
828
+ "Mosquito": 128,
829
+ "Motor vehicle (road)": 306,
830
+ "Motorboat, speedboat": 304,
831
+ "Motorcycle": 326,
832
+ "Mouse": 124,
833
+ "Music": 137,
834
+ "Music for children": 252,
835
+ "Music of Africa": 256,
836
+ "Music of Asia": 260,
837
+ "Music of Bollywood": 262,
838
+ "Music of Latin America": 248,
839
+ "Musical instrument": 138,
840
+ "Narration, monologue": 5,
841
+ "Neigh, whinny": 89,
842
+ "New-age music": 253,
843
+ "Noise": 513,
844
+ "Ocean": 294,
845
+ "Oink": 94,
846
+ "Opera": 238,
847
+ "Orchestra": 184,
848
+ "Organ": 155,
849
+ "Outside, rural or natural": 510,
850
+ "Outside, urban or manmade": 509,
851
+ "Owl": 119,
852
+ "Pant": 45,
853
+ "Patter": 125,
854
+ "Percussion": 161,
855
+ "Piano": 153,
856
+ "Pig": 93,
857
+ "Pigeon, dove": 115,
858
+ "Ping": 482,
859
+ "Pink noise": 521,
860
+ "Pizzicato": 192,
861
+ "Plop": 494,
862
+ "Plucked string instrument": 139,
863
+ "Police car (siren)": 323,
864
+ "Pop music": 216,
865
+ "Pour": 449,
866
+ "Power tool": 424,
867
+ "Power windows, electric windows": 311,
868
+ "Printer": 415,
869
+ "Progressive rock": 223,
870
+ "Propeller, airscrew": 338,
871
+ "Psychedelic rock": 225,
872
+ "Pulleys": 410,
873
+ "Pulse": 505,
874
+ "Pump (liquid)": 454,
875
+ "Punk rock": 221,
876
+ "Purr": 82,
877
+ "Quack": 105,
878
+ "Race car, auto racing": 315,
879
+ "Radio": 525,
880
+ "Rail transport": 328,
881
+ "Railroad car, train wagon": 332,
882
+ "Rain": 289,
883
+ "Rain on surface": 291,
884
+ "Raindrop": 290,
885
+ "Rapping": 36,
886
+ "Ratchet, pawl": 405,
887
+ "Rattle": 135,
888
+ "Rattle (instrument)": 175,
889
+ "Reggae": 228,
890
+ "Reverberation": 511,
891
+ "Reversing beeps": 319,
892
+ "Rhythm and blues": 226,
893
+ "Rimshot": 166,
894
+ "Ringtone": 391,
895
+ "Roar": 110,
896
+ "Roaring cats (lions, tigers)": 109,
897
+ "Rock and roll": 224,
898
+ "Rock music": 219,
899
+ "Rodents, rats, mice": 123,
900
+ "Roll": 477,
901
+ "Rowboat, canoe, kayak": 303,
902
+ "Rub": 476,
903
+ "Rumble": 493,
904
+ "Run": 51,
905
+ "Rustle": 487,
906
+ "Rustling leaves": 284,
907
+ "Sad music": 278,
908
+ "Sailboat, sailing ship": 302,
909
+ "Salsa music": 249,
910
+ "Sampler": 159,
911
+ "Sanding": 423,
912
+ "Sawing": 421,
913
+ "Saxophone": 197,
914
+ "Scary music": 282,
915
+ "Scissors": 381,
916
+ "Scrape": 475,
917
+ "Scratch": 474,
918
+ "Scratching (performance technique)": 215,
919
+ "Screaming": 14,
920
+ "Sewing machine": 411,
921
+ "Shatter": 443,
922
+ "Sheep": 97,
923
+ "Ship": 305,
924
+ "Shofar": 212,
925
+ "Shout": 8,
926
+ "Shuffle": 52,
927
+ "Shuffling cards": 383,
928
+ "Sidetone": 518,
929
+ "Sigh": 26,
930
+ "Silence": 500,
931
+ "Sine wave": 501,
932
+ "Singing": 27,
933
+ "Singing bowl": 214,
934
+ "Single-lens reflex camera": 417,
935
+ "Sink (filling or washing)": 371,
936
+ "Siren": 396,
937
+ "Sitar": 148,
938
+ "Sizzle": 490,
939
+ "Ska": 263,
940
+ "Skateboard": 342,
941
+ "Skidding": 312,
942
+ "Slam": 358,
943
+ "Slap, smack": 467,
944
+ "Sliding door": 357,
945
+ "Slosh": 446,
946
+ "Smash, crash": 469,
947
+ "Smoke detector, smoke alarm": 399,
948
+ "Snake": 134,
949
+ "Snare drum": 165,
950
+ "Sneeze": 49,
951
+ "Snicker": 19,
952
+ "Sniff": 50,
953
+ "Snoring": 43,
954
+ "Snort": 46,
955
+ "Sonar": 457,
956
+ "Song": 266,
957
+ "Soul music": 227,
958
+ "Sound effect": 504,
959
+ "Soundtrack music": 270,
960
+ "Speech": 0,
961
+ "Speech synthesizer": 7,
962
+ "Splash, splatter": 445,
963
+ "Splinter": 439,
964
+ "Spray": 453,
965
+ "Squawk": 114,
966
+ "Squeak": 361,
967
+ "Squeal": 485,
968
+ "Squish": 447,
969
+ "Static": 515,
970
+ "Steam": 296,
971
+ "Steam whistle": 403,
972
+ "Steel guitar, slide guitar": 144,
973
+ "Steelpan": 183,
974
+ "Stir": 455,
975
+ "Stomach rumble": 57,
976
+ "Stream": 292,
977
+ "String section": 190,
978
+ "Strum": 146,
979
+ "Subway, metro, underground": 334,
980
+ "Swing music": 230,
981
+ "Synthesizer": 158,
982
+ "Synthetic singing": 35,
983
+ "Tabla": 170,
984
+ "Tambourine": 174,
985
+ "Tap": 360,
986
+ "Tapping (guitar technique)": 145,
987
+ "Tearing": 480,
988
+ "Techno": 241,
989
+ "Telephone": 389,
990
+ "Telephone bell ringing": 390,
991
+ "Telephone dialing, DTMF": 392,
992
+ "Television": 524,
993
+ "Tender music": 279,
994
+ "Theme music": 268,
995
+ "Theremin": 213,
996
+ "Throat clearing": 48,
997
+ "Throbbing": 522,
998
+ "Thump, thud": 460,
999
+ "Thunder": 287,
1000
+ "Thunderstorm": 286,
1001
+ "Thunk": 461,
1002
+ "Tick": 407,
1003
+ "Tick-tock": 408,
1004
+ "Timpani": 169,
1005
+ "Tire squeal": 313,
1006
+ "Toilet flush": 374,
1007
+ "Tools": 418,
1008
+ "Toot": 309,
1009
+ "Toothbrush": 375,
1010
+ "Traditional music": 264,
1011
+ "Traffic noise, roadway noise": 327,
1012
+ "Train": 329,
1013
+ "Train horn": 331,
1014
+ "Train wheels squealing": 333,
1015
+ "Train whistle": 330,
1016
+ "Trance music": 247,
1017
+ "Trickle, dribble": 450,
1018
+ "Trombone": 188,
1019
+ "Truck": 316,
1020
+ "Trumpet": 187,
1021
+ "Tubular bells": 178,
1022
+ "Tuning fork": 204,
1023
+ "Turkey": 102,
1024
+ "Typewriter": 385,
1025
+ "Typing": 384,
1026
+ "Ukulele": 151,
1027
+ "Vacuum cleaner": 377,
1028
+ "Vehicle": 300,
1029
+ "Vehicle horn, car horn, honking": 308,
1030
+ "Vibraphone": 182,
1031
+ "Vibration": 523,
1032
+ "Video game music": 272,
1033
+ "Violin, fiddle": 191,
1034
+ "Vocal music": 254,
1035
+ "Wail, moan": 25,
1036
+ "Walk, footsteps": 53,
1037
+ "Water": 288,
1038
+ "Water tap, faucet": 370,
1039
+ "Waterfall": 293,
1040
+ "Waves, surf": 295,
1041
+ "Wedding music": 275,
1042
+ "Whack, thwack": 468,
1043
+ "Whale vocalization": 136,
1044
+ "Wheeze": 42,
1045
+ "Whimper": 24,
1046
+ "Whimper (dog)": 80,
1047
+ "Whip": 472,
1048
+ "Whir": 488,
1049
+ "Whispering": 15,
1050
+ "Whistle": 402,
1051
+ "Whistling": 40,
1052
+ "White noise": 520,
1053
+ "Whoop": 10,
1054
+ "Whoosh, swoosh, swish": 459,
1055
+ "Wild animals": 108,
1056
+ "Wind": 283,
1057
+ "Wind chime": 206,
1058
+ "Wind instrument, woodwind instrument": 195,
1059
+ "Wind noise (microphone)": 285,
1060
+ "Wood": 437,
1061
+ "Wood block": 173,
1062
+ "Writing": 387,
1063
+ "Yell": 11,
1064
+ "Yip": 76,
1065
+ "Yodeling": 29,
1066
+ "Zing": 497,
1067
+ "Zipper (clothing)": 378,
1068
+ "Zither": 150
1069
+ },
1070
+ "layer_norm_eps": 1e-12,
1071
+ "max_length": 1024,
1072
+ "model_type": "audio-spectrogram-transformer",
1073
+ "num_attention_heads": 3,
1074
+ "num_hidden_layers": 12,
1075
+ "num_mel_bins": 128,
1076
+ "patch_size": 16,
1077
+ "qkv_bias": true,
1078
+ "time_stride": 16,
1079
+ "torch_dtype": "float32",
1080
+ "transformers_version": "4.36.1"
1081
+ }
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2eb6aa49ea0881746707d22a9b30734cc7cf8b5dca3a228df0bef7246c2f8373
3
- size 21979944
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2bd1caae5db5d0495de86c332b3c75b12d6b2054669bfeb1bba13351f35e3fd6
3
+ size 22385252