nielsr HF staff commited on
Commit
3fd2720
1 Parent(s): 1cf70f9

Upload AudioSpectrogramTransformerForSequenceClassification

Browse files
Files changed (2) hide show
  1. config.json +1081 -0
  2. pytorch_model.bin +3 -0
config.json ADDED
@@ -0,0 +1,1081 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "AudioSpectrogramTransformerForSequenceClassification"
4
+ ],
5
+ "attention_probs_dropout_prob": 0.0,
6
+ "frequency_dimension": 128,
7
+ "frequency_stride": 14,
8
+ "hidden_act": "gelu",
9
+ "hidden_dropout_prob": 0.0,
10
+ "hidden_size": 768,
11
+ "id2label": {
12
+ "0": "Speech",
13
+ "1": "Male speech, man speaking",
14
+ "2": "Female speech, woman speaking",
15
+ "3": "Child speech, kid speaking",
16
+ "4": "Conversation",
17
+ "5": "Narration, monologue",
18
+ "6": "Babbling",
19
+ "7": "Speech synthesizer",
20
+ "8": "Shout",
21
+ "9": "Bellow",
22
+ "10": "Whoop",
23
+ "11": "Yell",
24
+ "12": "Battle cry",
25
+ "13": "Children shouting",
26
+ "14": "Screaming",
27
+ "15": "Whispering",
28
+ "16": "Laughter",
29
+ "17": "Baby laughter",
30
+ "18": "Giggle",
31
+ "19": "Snicker",
32
+ "20": "Belly laugh",
33
+ "21": "Chuckle, chortle",
34
+ "22": "Crying, sobbing",
35
+ "23": "Baby cry, infant cry",
36
+ "24": "Whimper",
37
+ "25": "Wail, moan",
38
+ "26": "Sigh",
39
+ "27": "Singing",
40
+ "28": "Choir",
41
+ "29": "Yodeling",
42
+ "30": "Chant",
43
+ "31": "Mantra",
44
+ "32": "Male singing",
45
+ "33": "Female singing",
46
+ "34": "Child singing",
47
+ "35": "Synthetic singing",
48
+ "36": "Rapping",
49
+ "37": "Humming",
50
+ "38": "Groan",
51
+ "39": "Grunt",
52
+ "40": "Whistling",
53
+ "41": "Breathing",
54
+ "42": "Wheeze",
55
+ "43": "Snoring",
56
+ "44": "Gasp",
57
+ "45": "Pant",
58
+ "46": "Snort",
59
+ "47": "Cough",
60
+ "48": "Throat clearing",
61
+ "49": "Sneeze",
62
+ "50": "Sniff",
63
+ "51": "Run",
64
+ "52": "Shuffle",
65
+ "53": "Walk, footsteps",
66
+ "54": "Chewing, mastication",
67
+ "55": "Biting",
68
+ "56": "Gargling",
69
+ "57": "Stomach rumble",
70
+ "58": "Burping, eructation",
71
+ "59": "Hiccup",
72
+ "60": "Fart",
73
+ "61": "Hands",
74
+ "62": "Finger snapping",
75
+ "63": "Clapping",
76
+ "64": "Heart sounds, heartbeat",
77
+ "65": "Heart murmur",
78
+ "66": "Cheering",
79
+ "67": "Applause",
80
+ "68": "Chatter",
81
+ "69": "Crowd",
82
+ "70": "Hubbub, speech noise, speech babble",
83
+ "71": "Children playing",
84
+ "72": "Animal",
85
+ "73": "Domestic animals, pets",
86
+ "74": "Dog",
87
+ "75": "Bark",
88
+ "76": "Yip",
89
+ "77": "Howl",
90
+ "78": "Bow-wow",
91
+ "79": "Growling",
92
+ "80": "Whimper (dog)",
93
+ "81": "Cat",
94
+ "82": "Purr",
95
+ "83": "Meow",
96
+ "84": "Hiss",
97
+ "85": "Caterwaul",
98
+ "86": "Livestock, farm animals, working animals",
99
+ "87": "Horse",
100
+ "88": "Clip-clop",
101
+ "89": "Neigh, whinny",
102
+ "90": "Cattle, bovinae",
103
+ "91": "Moo",
104
+ "92": "Cowbell",
105
+ "93": "Pig",
106
+ "94": "Oink",
107
+ "95": "Goat",
108
+ "96": "Bleat",
109
+ "97": "Sheep",
110
+ "98": "Fowl",
111
+ "99": "Chicken, rooster",
112
+ "100": "Cluck",
113
+ "101": "Crowing, cock-a-doodle-doo",
114
+ "102": "Turkey",
115
+ "103": "Gobble",
116
+ "104": "Duck",
117
+ "105": "Quack",
118
+ "106": "Goose",
119
+ "107": "Honk",
120
+ "108": "Wild animals",
121
+ "109": "Roaring cats (lions, tigers)",
122
+ "110": "Roar",
123
+ "111": "Bird",
124
+ "112": "Bird vocalization, bird call, bird song",
125
+ "113": "Chirp, tweet",
126
+ "114": "Squawk",
127
+ "115": "Pigeon, dove",
128
+ "116": "Coo",
129
+ "117": "Crow",
130
+ "118": "Caw",
131
+ "119": "Owl",
132
+ "120": "Hoot",
133
+ "121": "Bird flight, flapping wings",
134
+ "122": "Canidae, dogs, wolves",
135
+ "123": "Rodents, rats, mice",
136
+ "124": "Mouse",
137
+ "125": "Patter",
138
+ "126": "Insect",
139
+ "127": "Cricket",
140
+ "128": "Mosquito",
141
+ "129": "Fly, housefly",
142
+ "130": "Buzz",
143
+ "131": "Bee, wasp, etc.",
144
+ "132": "Frog",
145
+ "133": "Croak",
146
+ "134": "Snake",
147
+ "135": "Rattle",
148
+ "136": "Whale vocalization",
149
+ "137": "Music",
150
+ "138": "Musical instrument",
151
+ "139": "Plucked string instrument",
152
+ "140": "Guitar",
153
+ "141": "Electric guitar",
154
+ "142": "Bass guitar",
155
+ "143": "Acoustic guitar",
156
+ "144": "Steel guitar, slide guitar",
157
+ "145": "Tapping (guitar technique)",
158
+ "146": "Strum",
159
+ "147": "Banjo",
160
+ "148": "Sitar",
161
+ "149": "Mandolin",
162
+ "150": "Zither",
163
+ "151": "Ukulele",
164
+ "152": "Keyboard (musical)",
165
+ "153": "Piano",
166
+ "154": "Electric piano",
167
+ "155": "Organ",
168
+ "156": "Electronic organ",
169
+ "157": "Hammond organ",
170
+ "158": "Synthesizer",
171
+ "159": "Sampler",
172
+ "160": "Harpsichord",
173
+ "161": "Percussion",
174
+ "162": "Drum kit",
175
+ "163": "Drum machine",
176
+ "164": "Drum",
177
+ "165": "Snare drum",
178
+ "166": "Rimshot",
179
+ "167": "Drum roll",
180
+ "168": "Bass drum",
181
+ "169": "Timpani",
182
+ "170": "Tabla",
183
+ "171": "Cymbal",
184
+ "172": "Hi-hat",
185
+ "173": "Wood block",
186
+ "174": "Tambourine",
187
+ "175": "Rattle (instrument)",
188
+ "176": "Maraca",
189
+ "177": "Gong",
190
+ "178": "Tubular bells",
191
+ "179": "Mallet percussion",
192
+ "180": "Marimba, xylophone",
193
+ "181": "Glockenspiel",
194
+ "182": "Vibraphone",
195
+ "183": "Steelpan",
196
+ "184": "Orchestra",
197
+ "185": "Brass instrument",
198
+ "186": "French horn",
199
+ "187": "Trumpet",
200
+ "188": "Trombone",
201
+ "189": "Bowed string instrument",
202
+ "190": "String section",
203
+ "191": "Violin, fiddle",
204
+ "192": "Pizzicato",
205
+ "193": "Cello",
206
+ "194": "Double bass",
207
+ "195": "Wind instrument, woodwind instrument",
208
+ "196": "Flute",
209
+ "197": "Saxophone",
210
+ "198": "Clarinet",
211
+ "199": "Harp",
212
+ "200": "Bell",
213
+ "201": "Church bell",
214
+ "202": "Jingle bell",
215
+ "203": "Bicycle bell",
216
+ "204": "Tuning fork",
217
+ "205": "Chime",
218
+ "206": "Wind chime",
219
+ "207": "Change ringing (campanology)",
220
+ "208": "Harmonica",
221
+ "209": "Accordion",
222
+ "210": "Bagpipes",
223
+ "211": "Didgeridoo",
224
+ "212": "Shofar",
225
+ "213": "Theremin",
226
+ "214": "Singing bowl",
227
+ "215": "Scratching (performance technique)",
228
+ "216": "Pop music",
229
+ "217": "Hip hop music",
230
+ "218": "Beatboxing",
231
+ "219": "Rock music",
232
+ "220": "Heavy metal",
233
+ "221": "Punk rock",
234
+ "222": "Grunge",
235
+ "223": "Progressive rock",
236
+ "224": "Rock and roll",
237
+ "225": "Psychedelic rock",
238
+ "226": "Rhythm and blues",
239
+ "227": "Soul music",
240
+ "228": "Reggae",
241
+ "229": "Country",
242
+ "230": "Swing music",
243
+ "231": "Bluegrass",
244
+ "232": "Funk",
245
+ "233": "Folk music",
246
+ "234": "Middle Eastern music",
247
+ "235": "Jazz",
248
+ "236": "Disco",
249
+ "237": "Classical music",
250
+ "238": "Opera",
251
+ "239": "Electronic music",
252
+ "240": "House music",
253
+ "241": "Techno",
254
+ "242": "Dubstep",
255
+ "243": "Drum and bass",
256
+ "244": "Electronica",
257
+ "245": "Electronic dance music",
258
+ "246": "Ambient music",
259
+ "247": "Trance music",
260
+ "248": "Music of Latin America",
261
+ "249": "Salsa music",
262
+ "250": "Flamenco",
263
+ "251": "Blues",
264
+ "252": "Music for children",
265
+ "253": "New-age music",
266
+ "254": "Vocal music",
267
+ "255": "A capella",
268
+ "256": "Music of Africa",
269
+ "257": "Afrobeat",
270
+ "258": "Christian music",
271
+ "259": "Gospel music",
272
+ "260": "Music of Asia",
273
+ "261": "Carnatic music",
274
+ "262": "Music of Bollywood",
275
+ "263": "Ska",
276
+ "264": "Traditional music",
277
+ "265": "Independent music",
278
+ "266": "Song",
279
+ "267": "Background music",
280
+ "268": "Theme music",
281
+ "269": "Jingle (music)",
282
+ "270": "Soundtrack music",
283
+ "271": "Lullaby",
284
+ "272": "Video game music",
285
+ "273": "Christmas music",
286
+ "274": "Dance music",
287
+ "275": "Wedding music",
288
+ "276": "Happy music",
289
+ "277": "Funny music",
290
+ "278": "Sad music",
291
+ "279": "Tender music",
292
+ "280": "Exciting music",
293
+ "281": "Angry music",
294
+ "282": "Scary music",
295
+ "283": "Wind",
296
+ "284": "Rustling leaves",
297
+ "285": "Wind noise (microphone)",
298
+ "286": "Thunderstorm",
299
+ "287": "Thunder",
300
+ "288": "Water",
301
+ "289": "Rain",
302
+ "290": "Raindrop",
303
+ "291": "Rain on surface",
304
+ "292": "Stream",
305
+ "293": "Waterfall",
306
+ "294": "Ocean",
307
+ "295": "Waves, surf",
308
+ "296": "Steam",
309
+ "297": "Gurgling",
310
+ "298": "Fire",
311
+ "299": "Crackle",
312
+ "300": "Vehicle",
313
+ "301": "Boat, Water vehicle",
314
+ "302": "Sailboat, sailing ship",
315
+ "303": "Rowboat, canoe, kayak",
316
+ "304": "Motorboat, speedboat",
317
+ "305": "Ship",
318
+ "306": "Motor vehicle (road)",
319
+ "307": "Car",
320
+ "308": "Vehicle horn, car horn, honking",
321
+ "309": "Toot",
322
+ "310": "Car alarm",
323
+ "311": "Power windows, electric windows",
324
+ "312": "Skidding",
325
+ "313": "Tire squeal",
326
+ "314": "Car passing by",
327
+ "315": "Race car, auto racing",
328
+ "316": "Truck",
329
+ "317": "Air brake",
330
+ "318": "Air horn, truck horn",
331
+ "319": "Reversing beeps",
332
+ "320": "Ice cream truck, ice cream van",
333
+ "321": "Bus",
334
+ "322": "Emergency vehicle",
335
+ "323": "Police car (siren)",
336
+ "324": "Ambulance (siren)",
337
+ "325": "Fire engine, fire truck (siren)",
338
+ "326": "Motorcycle",
339
+ "327": "Traffic noise, roadway noise",
340
+ "328": "Rail transport",
341
+ "329": "Train",
342
+ "330": "Train whistle",
343
+ "331": "Train horn",
344
+ "332": "Railroad car, train wagon",
345
+ "333": "Train wheels squealing",
346
+ "334": "Subway, metro, underground",
347
+ "335": "Aircraft",
348
+ "336": "Aircraft engine",
349
+ "337": "Jet engine",
350
+ "338": "Propeller, airscrew",
351
+ "339": "Helicopter",
352
+ "340": "Fixed-wing aircraft, airplane",
353
+ "341": "Bicycle",
354
+ "342": "Skateboard",
355
+ "343": "Engine",
356
+ "344": "Light engine (high frequency)",
357
+ "345": "Dental drill, dentist's drill",
358
+ "346": "Lawn mower",
359
+ "347": "Chainsaw",
360
+ "348": "Medium engine (mid frequency)",
361
+ "349": "Heavy engine (low frequency)",
362
+ "350": "Engine knocking",
363
+ "351": "Engine starting",
364
+ "352": "Idling",
365
+ "353": "Accelerating, revving, vroom",
366
+ "354": "Door",
367
+ "355": "Doorbell",
368
+ "356": "Ding-dong",
369
+ "357": "Sliding door",
370
+ "358": "Slam",
371
+ "359": "Knock",
372
+ "360": "Tap",
373
+ "361": "Squeak",
374
+ "362": "Cupboard open or close",
375
+ "363": "Drawer open or close",
376
+ "364": "Dishes, pots, and pans",
377
+ "365": "Cutlery, silverware",
378
+ "366": "Chopping (food)",
379
+ "367": "Frying (food)",
380
+ "368": "Microwave oven",
381
+ "369": "Blender",
382
+ "370": "Water tap, faucet",
383
+ "371": "Sink (filling or washing)",
384
+ "372": "Bathtub (filling or washing)",
385
+ "373": "Hair dryer",
386
+ "374": "Toilet flush",
387
+ "375": "Toothbrush",
388
+ "376": "Electric toothbrush",
389
+ "377": "Vacuum cleaner",
390
+ "378": "Zipper (clothing)",
391
+ "379": "Keys jangling",
392
+ "380": "Coin (dropping)",
393
+ "381": "Scissors",
394
+ "382": "Electric shaver, electric razor",
395
+ "383": "Shuffling cards",
396
+ "384": "Typing",
397
+ "385": "Typewriter",
398
+ "386": "Computer keyboard",
399
+ "387": "Writing",
400
+ "388": "Alarm",
401
+ "389": "Telephone",
402
+ "390": "Telephone bell ringing",
403
+ "391": "Ringtone",
404
+ "392": "Telephone dialing, DTMF",
405
+ "393": "Dial tone",
406
+ "394": "Busy signal",
407
+ "395": "Alarm clock",
408
+ "396": "Siren",
409
+ "397": "Civil defense siren",
410
+ "398": "Buzzer",
411
+ "399": "Smoke detector, smoke alarm",
412
+ "400": "Fire alarm",
413
+ "401": "Foghorn",
414
+ "402": "Whistle",
415
+ "403": "Steam whistle",
416
+ "404": "Mechanisms",
417
+ "405": "Ratchet, pawl",
418
+ "406": "Clock",
419
+ "407": "Tick",
420
+ "408": "Tick-tock",
421
+ "409": "Gears",
422
+ "410": "Pulleys",
423
+ "411": "Sewing machine",
424
+ "412": "Mechanical fan",
425
+ "413": "Air conditioning",
426
+ "414": "Cash register",
427
+ "415": "Printer",
428
+ "416": "Camera",
429
+ "417": "Single-lens reflex camera",
430
+ "418": "Tools",
431
+ "419": "Hammer",
432
+ "420": "Jackhammer",
433
+ "421": "Sawing",
434
+ "422": "Filing (rasp)",
435
+ "423": "Sanding",
436
+ "424": "Power tool",
437
+ "425": "Drill",
438
+ "426": "Explosion",
439
+ "427": "Gunshot, gunfire",
440
+ "428": "Machine gun",
441
+ "429": "Fusillade",
442
+ "430": "Artillery fire",
443
+ "431": "Cap gun",
444
+ "432": "Fireworks",
445
+ "433": "Firecracker",
446
+ "434": "Burst, pop",
447
+ "435": "Eruption",
448
+ "436": "Boom",
449
+ "437": "Wood",
450
+ "438": "Chop",
451
+ "439": "Splinter",
452
+ "440": "Crack",
453
+ "441": "Glass",
454
+ "442": "Chink, clink",
455
+ "443": "Shatter",
456
+ "444": "Liquid",
457
+ "445": "Splash, splatter",
458
+ "446": "Slosh",
459
+ "447": "Squish",
460
+ "448": "Drip",
461
+ "449": "Pour",
462
+ "450": "Trickle, dribble",
463
+ "451": "Gush",
464
+ "452": "Fill (with liquid)",
465
+ "453": "Spray",
466
+ "454": "Pump (liquid)",
467
+ "455": "Stir",
468
+ "456": "Boiling",
469
+ "457": "Sonar",
470
+ "458": "Arrow",
471
+ "459": "Whoosh, swoosh, swish",
472
+ "460": "Thump, thud",
473
+ "461": "Thunk",
474
+ "462": "Electronic tuner",
475
+ "463": "Effects unit",
476
+ "464": "Chorus effect",
477
+ "465": "Basketball bounce",
478
+ "466": "Bang",
479
+ "467": "Slap, smack",
480
+ "468": "Whack, thwack",
481
+ "469": "Smash, crash",
482
+ "470": "Breaking",
483
+ "471": "Bouncing",
484
+ "472": "Whip",
485
+ "473": "Flap",
486
+ "474": "Scratch",
487
+ "475": "Scrape",
488
+ "476": "Rub",
489
+ "477": "Roll",
490
+ "478": "Crushing",
491
+ "479": "Crumpling, crinkling",
492
+ "480": "Tearing",
493
+ "481": "Beep, bleep",
494
+ "482": "Ping",
495
+ "483": "Ding",
496
+ "484": "Clang",
497
+ "485": "Squeal",
498
+ "486": "Creak",
499
+ "487": "Rustle",
500
+ "488": "Whir",
501
+ "489": "Clatter",
502
+ "490": "Sizzle",
503
+ "491": "Clicking",
504
+ "492": "Clickety-clack",
505
+ "493": "Rumble",
506
+ "494": "Plop",
507
+ "495": "Jingle, tinkle",
508
+ "496": "Hum",
509
+ "497": "Zing",
510
+ "498": "Boing",
511
+ "499": "Crunch",
512
+ "500": "Silence",
513
+ "501": "Sine wave",
514
+ "502": "Harmonic",
515
+ "503": "Chirp tone",
516
+ "504": "Sound effect",
517
+ "505": "Pulse",
518
+ "506": "Inside, small room",
519
+ "507": "Inside, large room or hall",
520
+ "508": "Inside, public space",
521
+ "509": "Outside, urban or manmade",
522
+ "510": "Outside, rural or natural",
523
+ "511": "Reverberation",
524
+ "512": "Echo",
525
+ "513": "Noise",
526
+ "514": "Environmental noise",
527
+ "515": "Static",
528
+ "516": "Mains hum",
529
+ "517": "Distortion",
530
+ "518": "Sidetone",
531
+ "519": "Cacophony",
532
+ "520": "White noise",
533
+ "521": "Pink noise",
534
+ "522": "Throbbing",
535
+ "523": "Vibration",
536
+ "524": "Television",
537
+ "525": "Radio",
538
+ "526": "Field recording"
539
+ },
540
+ "initializer_range": 0.02,
541
+ "intermediate_size": 3072,
542
+ "label2id": {
543
+ "A capella": 255,
544
+ "Accelerating, revving, vroom": 353,
545
+ "Accordion": 209,
546
+ "Acoustic guitar": 143,
547
+ "Afrobeat": 257,
548
+ "Air brake": 317,
549
+ "Air conditioning": 413,
550
+ "Air horn, truck horn": 318,
551
+ "Aircraft": 335,
552
+ "Aircraft engine": 336,
553
+ "Alarm": 388,
554
+ "Alarm clock": 395,
555
+ "Ambient music": 246,
556
+ "Ambulance (siren)": 324,
557
+ "Angry music": 281,
558
+ "Animal": 72,
559
+ "Applause": 67,
560
+ "Arrow": 458,
561
+ "Artillery fire": 430,
562
+ "Babbling": 6,
563
+ "Baby cry, infant cry": 23,
564
+ "Baby laughter": 17,
565
+ "Background music": 267,
566
+ "Bagpipes": 210,
567
+ "Bang": 466,
568
+ "Banjo": 147,
569
+ "Bark": 75,
570
+ "Basketball bounce": 465,
571
+ "Bass drum": 168,
572
+ "Bass guitar": 142,
573
+ "Bathtub (filling or washing)": 372,
574
+ "Battle cry": 12,
575
+ "Beatboxing": 218,
576
+ "Bee, wasp, etc.": 131,
577
+ "Beep, bleep": 481,
578
+ "Bell": 200,
579
+ "Bellow": 9,
580
+ "Belly laugh": 20,
581
+ "Bicycle": 341,
582
+ "Bicycle bell": 203,
583
+ "Bird": 111,
584
+ "Bird flight, flapping wings": 121,
585
+ "Bird vocalization, bird call, bird song": 112,
586
+ "Biting": 55,
587
+ "Bleat": 96,
588
+ "Blender": 369,
589
+ "Bluegrass": 231,
590
+ "Blues": 251,
591
+ "Boat, Water vehicle": 301,
592
+ "Boiling": 456,
593
+ "Boing": 498,
594
+ "Boom": 436,
595
+ "Bouncing": 471,
596
+ "Bow-wow": 78,
597
+ "Bowed string instrument": 189,
598
+ "Brass instrument": 185,
599
+ "Breaking": 470,
600
+ "Breathing": 41,
601
+ "Burping, eructation": 58,
602
+ "Burst, pop": 434,
603
+ "Bus": 321,
604
+ "Busy signal": 394,
605
+ "Buzz": 130,
606
+ "Buzzer": 398,
607
+ "Cacophony": 519,
608
+ "Camera": 416,
609
+ "Canidae, dogs, wolves": 122,
610
+ "Cap gun": 431,
611
+ "Car": 307,
612
+ "Car alarm": 310,
613
+ "Car passing by": 314,
614
+ "Carnatic music": 261,
615
+ "Cash register": 414,
616
+ "Cat": 81,
617
+ "Caterwaul": 85,
618
+ "Cattle, bovinae": 90,
619
+ "Caw": 118,
620
+ "Cello": 193,
621
+ "Chainsaw": 347,
622
+ "Change ringing (campanology)": 207,
623
+ "Chant": 30,
624
+ "Chatter": 68,
625
+ "Cheering": 66,
626
+ "Chewing, mastication": 54,
627
+ "Chicken, rooster": 99,
628
+ "Child singing": 34,
629
+ "Child speech, kid speaking": 3,
630
+ "Children playing": 71,
631
+ "Children shouting": 13,
632
+ "Chime": 205,
633
+ "Chink, clink": 442,
634
+ "Chirp tone": 503,
635
+ "Chirp, tweet": 113,
636
+ "Choir": 28,
637
+ "Chop": 438,
638
+ "Chopping (food)": 366,
639
+ "Chorus effect": 464,
640
+ "Christian music": 258,
641
+ "Christmas music": 273,
642
+ "Chuckle, chortle": 21,
643
+ "Church bell": 201,
644
+ "Civil defense siren": 397,
645
+ "Clang": 484,
646
+ "Clapping": 63,
647
+ "Clarinet": 198,
648
+ "Classical music": 237,
649
+ "Clatter": 489,
650
+ "Clickety-clack": 492,
651
+ "Clicking": 491,
652
+ "Clip-clop": 88,
653
+ "Clock": 406,
654
+ "Cluck": 100,
655
+ "Coin (dropping)": 380,
656
+ "Computer keyboard": 386,
657
+ "Conversation": 4,
658
+ "Coo": 116,
659
+ "Cough": 47,
660
+ "Country": 229,
661
+ "Cowbell": 92,
662
+ "Crack": 440,
663
+ "Crackle": 299,
664
+ "Creak": 486,
665
+ "Cricket": 127,
666
+ "Croak": 133,
667
+ "Crow": 117,
668
+ "Crowd": 69,
669
+ "Crowing, cock-a-doodle-doo": 101,
670
+ "Crumpling, crinkling": 479,
671
+ "Crunch": 499,
672
+ "Crushing": 478,
673
+ "Crying, sobbing": 22,
674
+ "Cupboard open or close": 362,
675
+ "Cutlery, silverware": 365,
676
+ "Cymbal": 171,
677
+ "Dance music": 274,
678
+ "Dental drill, dentist's drill": 345,
679
+ "Dial tone": 393,
680
+ "Didgeridoo": 211,
681
+ "Ding": 483,
682
+ "Ding-dong": 356,
683
+ "Disco": 236,
684
+ "Dishes, pots, and pans": 364,
685
+ "Distortion": 517,
686
+ "Dog": 74,
687
+ "Domestic animals, pets": 73,
688
+ "Door": 354,
689
+ "Doorbell": 355,
690
+ "Double bass": 194,
691
+ "Drawer open or close": 363,
692
+ "Drill": 425,
693
+ "Drip": 448,
694
+ "Drum": 164,
695
+ "Drum and bass": 243,
696
+ "Drum kit": 162,
697
+ "Drum machine": 163,
698
+ "Drum roll": 167,
699
+ "Dubstep": 242,
700
+ "Duck": 104,
701
+ "Echo": 512,
702
+ "Effects unit": 463,
703
+ "Electric guitar": 141,
704
+ "Electric piano": 154,
705
+ "Electric shaver, electric razor": 382,
706
+ "Electric toothbrush": 376,
707
+ "Electronic dance music": 245,
708
+ "Electronic music": 239,
709
+ "Electronic organ": 156,
710
+ "Electronic tuner": 462,
711
+ "Electronica": 244,
712
+ "Emergency vehicle": 322,
713
+ "Engine": 343,
714
+ "Engine knocking": 350,
715
+ "Engine starting": 351,
716
+ "Environmental noise": 514,
717
+ "Eruption": 435,
718
+ "Exciting music": 280,
719
+ "Explosion": 426,
720
+ "Fart": 60,
721
+ "Female singing": 33,
722
+ "Female speech, woman speaking": 2,
723
+ "Field recording": 526,
724
+ "Filing (rasp)": 422,
725
+ "Fill (with liquid)": 452,
726
+ "Finger snapping": 62,
727
+ "Fire": 298,
728
+ "Fire alarm": 400,
729
+ "Fire engine, fire truck (siren)": 325,
730
+ "Firecracker": 433,
731
+ "Fireworks": 432,
732
+ "Fixed-wing aircraft, airplane": 340,
733
+ "Flamenco": 250,
734
+ "Flap": 473,
735
+ "Flute": 196,
736
+ "Fly, housefly": 129,
737
+ "Foghorn": 401,
738
+ "Folk music": 233,
739
+ "Fowl": 98,
740
+ "French horn": 186,
741
+ "Frog": 132,
742
+ "Frying (food)": 367,
743
+ "Funk": 232,
744
+ "Funny music": 277,
745
+ "Fusillade": 429,
746
+ "Gargling": 56,
747
+ "Gasp": 44,
748
+ "Gears": 409,
749
+ "Giggle": 18,
750
+ "Glass": 441,
751
+ "Glockenspiel": 181,
752
+ "Goat": 95,
753
+ "Gobble": 103,
754
+ "Gong": 177,
755
+ "Goose": 106,
756
+ "Gospel music": 259,
757
+ "Groan": 38,
758
+ "Growling": 79,
759
+ "Grunge": 222,
760
+ "Grunt": 39,
761
+ "Guitar": 140,
762
+ "Gunshot, gunfire": 427,
763
+ "Gurgling": 297,
764
+ "Gush": 451,
765
+ "Hair dryer": 373,
766
+ "Hammer": 419,
767
+ "Hammond organ": 157,
768
+ "Hands": 61,
769
+ "Happy music": 276,
770
+ "Harmonic": 502,
771
+ "Harmonica": 208,
772
+ "Harp": 199,
773
+ "Harpsichord": 160,
774
+ "Heart murmur": 65,
775
+ "Heart sounds, heartbeat": 64,
776
+ "Heavy engine (low frequency)": 349,
777
+ "Heavy metal": 220,
778
+ "Helicopter": 339,
779
+ "Hi-hat": 172,
780
+ "Hiccup": 59,
781
+ "Hip hop music": 217,
782
+ "Hiss": 84,
783
+ "Honk": 107,
784
+ "Hoot": 120,
785
+ "Horse": 87,
786
+ "House music": 240,
787
+ "Howl": 77,
788
+ "Hubbub, speech noise, speech babble": 70,
789
+ "Hum": 496,
790
+ "Humming": 37,
791
+ "Ice cream truck, ice cream van": 320,
792
+ "Idling": 352,
793
+ "Independent music": 265,
794
+ "Insect": 126,
795
+ "Inside, large room or hall": 507,
796
+ "Inside, public space": 508,
797
+ "Inside, small room": 506,
798
+ "Jackhammer": 420,
799
+ "Jazz": 235,
800
+ "Jet engine": 337,
801
+ "Jingle (music)": 269,
802
+ "Jingle bell": 202,
803
+ "Jingle, tinkle": 495,
804
+ "Keyboard (musical)": 152,
805
+ "Keys jangling": 379,
806
+ "Knock": 359,
807
+ "Laughter": 16,
808
+ "Lawn mower": 346,
809
+ "Light engine (high frequency)": 344,
810
+ "Liquid": 444,
811
+ "Livestock, farm animals, working animals": 86,
812
+ "Lullaby": 271,
813
+ "Machine gun": 428,
814
+ "Mains hum": 516,
815
+ "Male singing": 32,
816
+ "Male speech, man speaking": 1,
817
+ "Mallet percussion": 179,
818
+ "Mandolin": 149,
819
+ "Mantra": 31,
820
+ "Maraca": 176,
821
+ "Marimba, xylophone": 180,
822
+ "Mechanical fan": 412,
823
+ "Mechanisms": 404,
824
+ "Medium engine (mid frequency)": 348,
825
+ "Meow": 83,
826
+ "Microwave oven": 368,
827
+ "Middle Eastern music": 234,
828
+ "Moo": 91,
829
+ "Mosquito": 128,
830
+ "Motor vehicle (road)": 306,
831
+ "Motorboat, speedboat": 304,
832
+ "Motorcycle": 326,
833
+ "Mouse": 124,
834
+ "Music": 137,
835
+ "Music for children": 252,
836
+ "Music of Africa": 256,
837
+ "Music of Asia": 260,
838
+ "Music of Bollywood": 262,
839
+ "Music of Latin America": 248,
840
+ "Musical instrument": 138,
841
+ "Narration, monologue": 5,
842
+ "Neigh, whinny": 89,
843
+ "New-age music": 253,
844
+ "Noise": 513,
845
+ "Ocean": 294,
846
+ "Oink": 94,
847
+ "Opera": 238,
848
+ "Orchestra": 184,
849
+ "Organ": 155,
850
+ "Outside, rural or natural": 510,
851
+ "Outside, urban or manmade": 509,
852
+ "Owl": 119,
853
+ "Pant": 45,
854
+ "Patter": 125,
855
+ "Percussion": 161,
856
+ "Piano": 153,
857
+ "Pig": 93,
858
+ "Pigeon, dove": 115,
859
+ "Ping": 482,
860
+ "Pink noise": 521,
861
+ "Pizzicato": 192,
862
+ "Plop": 494,
863
+ "Plucked string instrument": 139,
864
+ "Police car (siren)": 323,
865
+ "Pop music": 216,
866
+ "Pour": 449,
867
+ "Power tool": 424,
868
+ "Power windows, electric windows": 311,
869
+ "Printer": 415,
870
+ "Progressive rock": 223,
871
+ "Propeller, airscrew": 338,
872
+ "Psychedelic rock": 225,
873
+ "Pulleys": 410,
874
+ "Pulse": 505,
875
+ "Pump (liquid)": 454,
876
+ "Punk rock": 221,
877
+ "Purr": 82,
878
+ "Quack": 105,
879
+ "Race car, auto racing": 315,
880
+ "Radio": 525,
881
+ "Rail transport": 328,
882
+ "Railroad car, train wagon": 332,
883
+ "Rain": 289,
884
+ "Rain on surface": 291,
885
+ "Raindrop": 290,
886
+ "Rapping": 36,
887
+ "Ratchet, pawl": 405,
888
+ "Rattle": 135,
889
+ "Rattle (instrument)": 175,
890
+ "Reggae": 228,
891
+ "Reverberation": 511,
892
+ "Reversing beeps": 319,
893
+ "Rhythm and blues": 226,
894
+ "Rimshot": 166,
895
+ "Ringtone": 391,
896
+ "Roar": 110,
897
+ "Roaring cats (lions, tigers)": 109,
898
+ "Rock and roll": 224,
899
+ "Rock music": 219,
900
+ "Rodents, rats, mice": 123,
901
+ "Roll": 477,
902
+ "Rowboat, canoe, kayak": 303,
903
+ "Rub": 476,
904
+ "Rumble": 493,
905
+ "Run": 51,
906
+ "Rustle": 487,
907
+ "Rustling leaves": 284,
908
+ "Sad music": 278,
909
+ "Sailboat, sailing ship": 302,
910
+ "Salsa music": 249,
911
+ "Sampler": 159,
912
+ "Sanding": 423,
913
+ "Sawing": 421,
914
+ "Saxophone": 197,
915
+ "Scary music": 282,
916
+ "Scissors": 381,
917
+ "Scrape": 475,
918
+ "Scratch": 474,
919
+ "Scratching (performance technique)": 215,
920
+ "Screaming": 14,
921
+ "Sewing machine": 411,
922
+ "Shatter": 443,
923
+ "Sheep": 97,
924
+ "Ship": 305,
925
+ "Shofar": 212,
926
+ "Shout": 8,
927
+ "Shuffle": 52,
928
+ "Shuffling cards": 383,
929
+ "Sidetone": 518,
930
+ "Sigh": 26,
931
+ "Silence": 500,
932
+ "Sine wave": 501,
933
+ "Singing": 27,
934
+ "Singing bowl": 214,
935
+ "Single-lens reflex camera": 417,
936
+ "Sink (filling or washing)": 371,
937
+ "Siren": 396,
938
+ "Sitar": 148,
939
+ "Sizzle": 490,
940
+ "Ska": 263,
941
+ "Skateboard": 342,
942
+ "Skidding": 312,
943
+ "Slam": 358,
944
+ "Slap, smack": 467,
945
+ "Sliding door": 357,
946
+ "Slosh": 446,
947
+ "Smash, crash": 469,
948
+ "Smoke detector, smoke alarm": 399,
949
+ "Snake": 134,
950
+ "Snare drum": 165,
951
+ "Sneeze": 49,
952
+ "Snicker": 19,
953
+ "Sniff": 50,
954
+ "Snoring": 43,
955
+ "Snort": 46,
956
+ "Sonar": 457,
957
+ "Song": 266,
958
+ "Soul music": 227,
959
+ "Sound effect": 504,
960
+ "Soundtrack music": 270,
961
+ "Speech": 0,
962
+ "Speech synthesizer": 7,
963
+ "Splash, splatter": 445,
964
+ "Splinter": 439,
965
+ "Spray": 453,
966
+ "Squawk": 114,
967
+ "Squeak": 361,
968
+ "Squeal": 485,
969
+ "Squish": 447,
970
+ "Static": 515,
971
+ "Steam": 296,
972
+ "Steam whistle": 403,
973
+ "Steel guitar, slide guitar": 144,
974
+ "Steelpan": 183,
975
+ "Stir": 455,
976
+ "Stomach rumble": 57,
977
+ "Stream": 292,
978
+ "String section": 190,
979
+ "Strum": 146,
980
+ "Subway, metro, underground": 334,
981
+ "Swing music": 230,
982
+ "Synthesizer": 158,
983
+ "Synthetic singing": 35,
984
+ "Tabla": 170,
985
+ "Tambourine": 174,
986
+ "Tap": 360,
987
+ "Tapping (guitar technique)": 145,
988
+ "Tearing": 480,
989
+ "Techno": 241,
990
+ "Telephone": 389,
991
+ "Telephone bell ringing": 390,
992
+ "Telephone dialing, DTMF": 392,
993
+ "Television": 524,
994
+ "Tender music": 279,
995
+ "Theme music": 268,
996
+ "Theremin": 213,
997
+ "Throat clearing": 48,
998
+ "Throbbing": 522,
999
+ "Thump, thud": 460,
1000
+ "Thunder": 287,
1001
+ "Thunderstorm": 286,
1002
+ "Thunk": 461,
1003
+ "Tick": 407,
1004
+ "Tick-tock": 408,
1005
+ "Timpani": 169,
1006
+ "Tire squeal": 313,
1007
+ "Toilet flush": 374,
1008
+ "Tools": 418,
1009
+ "Toot": 309,
1010
+ "Toothbrush": 375,
1011
+ "Traditional music": 264,
1012
+ "Traffic noise, roadway noise": 327,
1013
+ "Train": 329,
1014
+ "Train horn": 331,
1015
+ "Train wheels squealing": 333,
1016
+ "Train whistle": 330,
1017
+ "Trance music": 247,
1018
+ "Trickle, dribble": 450,
1019
+ "Trombone": 188,
1020
+ "Truck": 316,
1021
+ "Trumpet": 187,
1022
+ "Tubular bells": 178,
1023
+ "Tuning fork": 204,
1024
+ "Turkey": 102,
1025
+ "Typewriter": 385,
1026
+ "Typing": 384,
1027
+ "Ukulele": 151,
1028
+ "Vacuum cleaner": 377,
1029
+ "Vehicle": 300,
1030
+ "Vehicle horn, car horn, honking": 308,
1031
+ "Vibraphone": 182,
1032
+ "Vibration": 523,
1033
+ "Video game music": 272,
1034
+ "Violin, fiddle": 191,
1035
+ "Vocal music": 254,
1036
+ "Wail, moan": 25,
1037
+ "Walk, footsteps": 53,
1038
+ "Water": 288,
1039
+ "Water tap, faucet": 370,
1040
+ "Waterfall": 293,
1041
+ "Waves, surf": 295,
1042
+ "Wedding music": 275,
1043
+ "Whack, thwack": 468,
1044
+ "Whale vocalization": 136,
1045
+ "Wheeze": 42,
1046
+ "Whimper": 24,
1047
+ "Whimper (dog)": 80,
1048
+ "Whip": 472,
1049
+ "Whir": 488,
1050
+ "Whispering": 15,
1051
+ "Whistle": 402,
1052
+ "Whistling": 40,
1053
+ "White noise": 520,
1054
+ "Whoop": 10,
1055
+ "Whoosh, swoosh, swish": 459,
1056
+ "Wild animals": 108,
1057
+ "Wind": 283,
1058
+ "Wind chime": 206,
1059
+ "Wind instrument, woodwind instrument": 195,
1060
+ "Wind noise (microphone)": 285,
1061
+ "Wood": 437,
1062
+ "Wood block": 173,
1063
+ "Writing": 387,
1064
+ "Yell": 11,
1065
+ "Yip": 76,
1066
+ "Yodeling": 29,
1067
+ "Zing": 497,
1068
+ "Zipper (clothing)": 378,
1069
+ "Zither": 150
1070
+ },
1071
+ "layer_norm_eps": 1e-12,
1072
+ "model_type": "audio-spectrogram-transformer",
1073
+ "num_attention_heads": 12,
1074
+ "num_hidden_layers": 12,
1075
+ "patch_size": 16,
1076
+ "qkv_bias": true,
1077
+ "time_dimension": 1024,
1078
+ "time_stride": 14,
1079
+ "torch_dtype": "float32",
1080
+ "transformers_version": "4.25.0.dev0"
1081
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0bcbd617aa2913004b0d33d5d86c6667863666e09ec4f3be2e5f1224e9d7c0b5
3
+ size 344740651