Hugging Face's logo

Datasets:
zed-m97
/
nano4m-Audio

Tasks:
Audio Classification
Text-to-Audio
Audio-to-Audio
Modalities:
Text
Audio
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
audio
multimodal
nano4m
4m
vggsound
encodec
License:
Dataset card Data Studio Files
xet
 Community
1
zed-m97 commited on
Commit
bcfeb3d
·
verified ·
1 Parent(s): de0857e

per-class stats + score histograms

Browse files
Files changed (1) hide show
  1. stats_person2.md +161 -0
stats_person2.md ADDED
@@ -0,0 +1,161 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Person-2 dataset stats
2
+
3
+ Manifest: `data/manifest_person2.csv` (2155 rows)
4
+
5
+ ## Status breakdown
6
+
7
+ | status | count | % of total |
8
+ |---|---:|---:|
9
+ | `OK` | 1729 | 80.2% |
10
+ | `FAIL_DL` | 385 | 17.9% |
11
+ | `FAIL_CLIP` | 26 | 1.2% |
12
+ | `FAIL_CLAP` | 14 | 0.6% |
13
+ | `FAIL_PEAKS` | 1 | 0.0% |
14
+
15
+ ## Per-class survival (OK / total)
16
+
17
+ | class | train OK | test OK | total OK | total raw | survival |
18
+ |---|---:|---:|---:|---:|---:|
19
+ | cat meowing | 449 | 41 | 490 | 588 | 83.3% |
20
+ | coyote howling | 344 | 41 | 385 | 490 | 78.6% |
21
+ | dog barking | 599 | 38 | 637 | 767 | 83.1% |
22
+ | elephant trumpeting | 183 | 34 | 217 | 310 | 70.0% |
23
+
24
+ ## Failure-mode breakdown per class
25
+
26
+ | class | FAIL_DL | FAIL_CLIP | FAIL_CLAP | FAIL_PEAKS | MISSING |
27
+ |---|---:|---:|---:|---:|---:|
28
+ | cat meowing | 88 | 6 | 4 | 0 | 0 |
29
+ | coyote howling | 90 | 9 | 5 | 1 | 0 |
30
+ | dog barking | 127 | 3 | 0 | 0 | 0 |
31
+ | elephant trumpeting | 80 | 8 | 5 | 0 | 0 |
32
+
33
+ ## CLIP score distribution (OK only)
34
+
35
+ Non-NaN OK rows: 1720 / 1729 (NaN = smoke-test residuals predating canonical log)
36
+
37
+ threshold = 0.2, mean = 0.283, std = 0.025, min = 0.200, max = 0.366
38
+
39
+ ```
40
+ [0.200, 0.211) 20 ##
41
+ [0.211, 0.222) 18 ##
42
+ [0.222, 0.233) 38 ####
43
+ [0.233, 0.244) 45 #####
44
+ [0.244, 0.255) 82 #########
45
+ [0.255, 0.266) 159 ##################
46
+ [0.266, 0.277) 283 #################################
47
+ [0.277, 0.289) 338 #######################################
48
+ [0.289, 0.300) 346 ########################################
49
+ [0.300, 0.311) 204 ########################
50
+ [0.311, 0.322) 122 ##############
51
+ [0.322, 0.333) 26 ###
52
+ [0.333, 0.344) 25 ###
53
+ [0.344, 0.355) 11 #
54
+ [0.355, 0.366) 3
55
+ ```
56
+
57
+ Per-class CLIP mean ± std:
58
+
59
+ | class | mean | std | min | max |
60
+ |---|---:|---:|---:|---:|
61
+ | cat meowing | 0.275 | 0.020 | 0.203 | 0.329 |
62
+ | coyote howling | 0.284 | 0.036 | 0.200 | 0.366 |
63
+ | dog barking | 0.288 | 0.018 | 0.201 | 0.347 |
64
+ | elephant trumpeting | 0.284 | 0.024 | 0.202 | 0.334 |
65
+
66
+ ## CLAP score distribution (OK only, raw logits)
67
+
68
+ Non-NaN OK rows: 1720 / 1729
69
+
70
+ threshold = 0.2, mean = 11.71, std = 3.02, min = 0.28, max = 20.18
71
+
72
+ ```
73
+ [ 0.28, 1.60) 16 ##
74
+ [ 1.60, 2.93) 13 #
75
+ [ 2.93, 4.26) 27 ###
76
+ [ 4.26, 5.58) 36 ###
77
+ [ 5.58, 6.91) 42 ####
78
+ [ 6.91, 8.24) 64 ######
79
+ [ 8.24, 9.56) 105 ##########
80
+ [ 9.56, 10.89) 218 ####################
81
+ [ 10.89, 12.22) 347 #################################
82
+ [ 12.22, 13.54) 426 ########################################
83
+ [ 13.54, 14.87) 235 ######################
84
+ [ 14.87, 16.20) 122 ###########
85
+ [ 16.20, 17.52) 53 #####
86
+ [ 17.52, 18.85) 15 #
87
+ [ 18.85, 20.18) 1
88
+ ```
89
+
90
+ Per-class CLAP mean ± std:
91
+
92
+ | class | mean | std | min | max |
93
+ |---|---:|---:|---:|---:|
94
+ | cat meowing | 12.03 | 2.83 | 0.72 | 18.77 |
95
+ | coyote howling | 10.63 | 3.30 | 0.28 | 17.08 |
96
+ | dog barking | 12.48 | 2.01 | 3.36 | 18.70 |
97
+ | elephant trumpeting | 10.64 | 4.32 | 0.52 | 20.18 |
98
+
99
+ ## Audio sample-count uniformity
100
+
101
+ unique sample counts: 1 (target = 1, all clips at 41,040 = 1.7100 s @ 24 kHz)
102
+ 41040 samples (1.7100 s): 1729 clips
103
+
104
+ ## Qualitative samples (3 best + 3 worst CLIP score per class)
105
+
106
+ Copied to `data/_qualitative/{class}/{best,worst}_clip/{stem}.{wav,jpg}` for eyeball/listen.
107
+
108
+ ### cat meowing
109
+
110
+ | rank | stem | split | clip | clap |
111
+ |---|---|---|---:|---:|
112
+ | best #1 | `eo2HskB5Y8I_000015` | train | 0.329 | 10.11 |
113
+ | best #2 | `rjm_phhaTqg_000012` | train | 0.325 | 15.27 |
114
+ | best #3 | `84joiXni4uk_000030` | train | 0.321 | 13.49 |
115
+ | worst #1 | `18HRNQyo9mg_000025` | test | 0.203 | 6.14 |
116
+ | worst #2 | `_V24O52TldA_000004` | train | 0.204 | 14.73 |
117
+ | worst #3 | `1b1b-9_AcFg_000030` | train | 0.205 | 14.53 |
118
+
119
+ ### coyote howling
120
+
121
+ | rank | stem | split | clip | clap |
122
+ |---|---|---|---:|---:|
123
+ | best #1 | `1J5LA2V-68k_000011` | train | 0.366 | 8.46 |
124
+ | best #2 | `1J5LA2V-68k_000043` | train | 0.366 | 11.50 |
125
+ | best #3 | `XA6RBioito4_000067` | test | 0.364 | 9.92 |
126
+ | worst #1 | `1Jb2bUPCU8U_000056` | train | 0.200 | 13.40 |
127
+ | worst #2 | `hRphojIYqiw_000018` | train | 0.201 | 8.60 |
128
+ | worst #3 | `WKbADRKJLo0_000138` | test | 0.205 | 9.26 |
129
+
130
+ ### dog barking
131
+
132
+ | rank | stem | split | clip | clap |
133
+ |---|---|---|---:|---:|
134
+ | best #1 | `urmwibHFOHc_000000` | train | 0.347 | 12.95 |
135
+ | best #2 | `wqMoQkNqb3U_000000` | train | 0.323 | 9.65 |
136
+ | best #3 | `VxsdrQ7a5eg_000000` | train | 0.322 | 11.61 |
137
+ | worst #1 | `UNzasfgSrTo_000010` | train | 0.201 | 12.46 |
138
+ | worst #2 | `LHjIPTqS94k_000030` | train | 0.206 | 18.42 |
139
+ | worst #3 | `m98BKPAD-84_000020` | test | 0.217 | 13.39 |
140
+
141
+ ### elephant trumpeting
142
+
143
+ | rank | stem | split | clip | clap |
144
+ |---|---|---|---:|---:|
145
+ | best #1 | `tY0ba7snoek_000005` | train | 0.334 | 16.31 |
146
+ | best #2 | `fy3-UPb_z4k_000017` | train | 0.332 | 15.09 |
147
+ | best #3 | `itTQe0xQUyw_000019` | train | 0.329 | 13.86 |
148
+ | worst #1 | `r-larRRhtco_000000` | train | 0.202 | 15.77 |
149
+ | worst #2 | `r-larRRhtco_000011` | train | 0.204 | 15.60 |
150
+ | worst #3 | `lEDHRh8gfm8_000094` | train | 0.205 | 3.16 |
151
+
152
+ ## Modality presence across OK rows
153
+
154
+ | modality | present / total | % |
155
+ |---|---:|---:|
156
+ | raw audio (wav) | 1729 / 1729 | 100.0% |
157
+ | raw image (jpg) | 1729 / 1729 | 100.0% |
158
+ | EnCodec tokens (tok_audio@256) | 1729 / 1729 | 100.0% |
159
+ | scene_desc JSON | 1729 / 1729 | 100.0% |
160
+ | Cosmos RGB tokens (tok_rgb@256, GPU step) | 0 / 1729 | 0.0% |
161
+