File size: 10,408 Bytes
605474c
fe31efd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
605474c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Engine Sound Classifier</title>
<style>
  :root {
    color-scheme: light;
    --page:        #f9f9f7;
    --surface-1:   #fcfcfb;
    --text-primary:   #0b0b0b;
    --text-secondary: #52514e;
    --text-muted:     #898781;
    --grid:        #e1e0d9;
    --baseline:    #c3c2b7;
    --border:      rgba(11,11,11,0.10);
    --series-1:    #2a78d6;
    --series-1-wash: rgba(42,120,214,0.10);
    --link:        #1c5cab;
  }
  @media (prefers-color-scheme: dark) {
    :root:where(:not([data-theme="light"])) {
      color-scheme: dark;
      --page:        #0d0d0d;
      --surface-1:   #1a1a19;
      --text-primary:   #ffffff;
      --text-secondary: #c3c2b7;
      --text-muted:     #898781;
      --grid:        #2c2c2a;
      --baseline:    #383835;
      --border:      rgba(255,255,255,0.10);
      --series-1:    #3987e5;
      --series-1-wash: rgba(57,135,229,0.14);
      --link:        #86b6ef;
    }
  }
  :root[data-theme="dark"] {
    color-scheme: dark;
    --page:        #0d0d0d;
    --surface-1:   #1a1a19;
    --text-primary:   #ffffff;
    --text-secondary: #c3c2b7;
    --text-muted:     #898781;
    --grid:        #2c2c2a;
    --baseline:    #383835;
    --border:      rgba(255,255,255,0.10);
    --series-1:    #3987e5;
    --series-1-wash: rgba(57,135,229,0.14);
    --link:        #86b6ef;
  }

  * { box-sizing: border-box; }
  body {
    margin: 0;
    background: var(--page);
    color: var(--text-primary);
    font: 16px/1.55 system-ui, -apple-system, "Segoe UI", sans-serif;
  }
  main {
    max-width: 760px;
    margin: 0 auto;
    padding: 48px 20px 80px;
  }
  h1 { font-size: 1.7rem; margin: 0 0 6px; }
  h2 { font-size: 1.15rem; margin: 40px 0 12px; }
  .tagline { color: var(--text-secondary); margin: 0 0 20px; }
  .status {
    display: inline-block;
    font-size: 0.85rem;
    color: var(--text-secondary);
    background: var(--series-1-wash);
    border: 1px solid var(--border);
    border-radius: 6px;
    padding: 4px 10px;
    margin-bottom: 20px;
  }
  p { color: var(--text-secondary); }
  a { color: var(--link); }
  code {
    background: var(--surface-1);
    border: 1px solid var(--border);
    border-radius: 4px;
    padding: 1px 5px;
    font-size: 0.9em;
  }

  .card {
    background: var(--surface-1);
    border: 1px solid var(--border);
    border-radius: 10px;
    padding: 20px;
  }

  .samples { display: grid; gap: 12px; }
  .sample-row {
    display: flex;
    align-items: center;
    gap: 14px;
    flex-wrap: wrap;
  }
  .sample-label { min-width: 170px; font-size: 0.95rem; color: var(--text-primary); }
  audio { flex: 1; min-width: 220px; height: 36px; }

  /* --- bar chart --- */
  .chart { padding: 4px 4px 0; }
  .bar-row {
    display: grid;
    grid-template-columns: 90px 1fr 56px;
    align-items: center;
    gap: 10px;
    height: 40px;
    position: relative;
  }
  .bar-row + .bar-row { border-top: 1px solid var(--grid); }
  .bar-cat { font-size: 0.9rem; color: var(--text-secondary); }
  .bar-track { position: relative; height: 24px; }
  .bar-fill {
    position: absolute;
    left: 0; top: 0;
    height: 24px;
    border-radius: 4px;
    background: var(--series-1);
    transition: filter 0.1s ease;
  }
  .bar-fill.best { background: var(--series-1); }
  .bar-fill:hover { filter: brightness(1.08); }
  .bar-value { font-size: 0.9rem; color: var(--text-primary); text-align: right; font-variant-numeric: tabular-nums; }
  .chart-axis {
    display: grid;
    grid-template-columns: 90px 1fr 56px;
    gap: 10px;
    margin-top: 6px;
    border-top: 1px solid var(--baseline);
    padding-top: 4px;
  }
  .chart-axis span { font-size: 0.75rem; color: var(--text-muted); }
  .chart-axis .ticks { position: relative; height: 14px; }
  .chart-axis .ticks i { position: absolute; top: 0; font-style: normal; }

  .tooltip {
    position: absolute;
    pointer-events: none;
    background: var(--text-primary);
    color: var(--page);
    font-size: 0.78rem;
    padding: 5px 8px;
    border-radius: 5px;
    white-space: nowrap;
    transform: translate(-50%, -130%);
    opacity: 0;
    transition: opacity 0.1s ease;
    z-index: 5;
  }
  .bar-row:hover .tooltip { opacity: 1; }

  table.perf {
    width: 100%;
    border-collapse: collapse;
    font-size: 0.88rem;
    margin-top: 14px;
  }
  table.perf caption { text-align: left; color: var(--text-muted); font-size: 0.78rem; margin-bottom: 6px; }
  table.perf th, table.perf td {
    text-align: left;
    padding: 6px 8px;
    border-bottom: 1px solid var(--grid);
    color: var(--text-secondary);
  }
  table.perf th { color: var(--text-primary); font-weight: 600; }
  table.perf td.num, table.perf th.num { text-align: right; font-variant-numeric: tabular-nums; }

  ul.weak { color: var(--text-secondary); padding-left: 20px; }
  ul.weak li { margin-bottom: 6px; }

  .links { display: flex; flex-wrap: wrap; gap: 10px 24px; margin-top: 8px; }
  .links a { font-size: 0.95rem; }

  footer {
    margin-top: 48px;
    padding-top: 16px;
    border-top: 1px solid var(--grid);
    color: var(--text-muted);
    font-size: 0.82rem;
  }
</style>
</head>
<body>
<main>
  <h1>Engine Sound Classifier</h1>
  <p class="tagline">
    A fine-tuned <a href="https://github.com/qiuqiangkong/audioset_tagging_cnn" target="_blank" rel="noopener">PANNs CNN14</a>
    (pretrained on AudioSet) that classifies short engine-audio clips into one of
    34 configurations — cylinder count, layout, and stroke type (e.g. <code>i4</code>,
    <code>v8_cross</code>, <code>single_two_stroke</code>, <code>2_rotor</code>).
  </p>
  <span class="status">Early checkpoint · ~0.33 balanced accuracy · 34 classes</span>

  <p>
    This is a static, informational Space — it doesn't run the model. It exists
    so you can hear a few labeled example clips and see how the model is doing
    without needing to spin up compute. To actually run inference, download the
    checkpoint from the model repo linked below.
  </p>

  <h2>Example clips</h2>
  <div class="card samples">
    <div class="sample-row">
      <span class="sample-label"><code>i2_360</code></span>
      <audio controls preload="none" src="audio/i2_360.wav"></audio>
    </div>
    <div class="sample-row">
      <span class="sample-label"><code>v2_45</code></span>
      <audio controls preload="none" src="audio/v2_45.wav"></audio>
    </div>
    <div class="sample-row">
      <span class="sample-label"><code>single_two_stroke</code></span>
      <audio controls preload="none" src="audio/single_two_stroke.wav"></audio>
    </div>
  </div>

  <h2>Performance by run</h2>
  <div class="card">
    <div class="chart" id="chart">
      <div class="bar-row">
        <span class="bar-cat">Run A</span>
        <div class="bar-track">
          <span style="font-size: 0.82rem; color: var(--text-muted);">balanced accuracy not tracked this run (micro 0.380, overfit)</span>
        </div>
        <span class="bar-value"></span>
      </div>
      <div class="bar-row">
        <span class="bar-cat">Run B</span>
        <div class="bar-track">
          <div class="bar-fill" style="width: 74.25%"></div>
          <div class="tooltip" style="left: 74.25%">0.297 balanced · 0.311 micro</div>
        </div>
        <span class="bar-value">0.297</span>
      </div>
      <div class="bar-row">
        <span class="bar-cat">Run C</span>
        <div class="bar-track">
          <div class="bar-fill" style="width: 80.75%"></div>
          <div class="tooltip" style="left: 80.75%">0.323 balanced · 0.343 micro</div>
        </div>
        <span class="bar-value">0.323</span>
      </div>
      <div class="bar-row">
        <span class="bar-cat">Run D</span>
        <div class="bar-track">
          <div class="bar-fill best" style="width: 82.5%"></div>
          <div class="tooltip" style="left: 82.5%">0.330 balanced · 0.376 micro (best)</div>
        </div>
        <span class="bar-value"><strong>0.330</strong></span>
      </div>
      <div class="chart-axis">
        <span></span>
        <div class="ticks">
          <i style="left: 0%">0.0</i>
          <i style="left: 50%">0.2</i>
          <i style="right: 0%">0.4</i>
        </div>
        <span></span>
      </div>
    </div>
    <table class="perf">
      <caption>Recording-level balanced accuracy is the headline metric; random guessing over 34 classes is ~0.03.</caption>
      <thead>
        <tr><th>Run</th><th>Change</th><th class="num">Balanced acc</th><th class="num">Micro acc</th></tr>
      </thead>
      <tbody>
        <tr><td>A</td><td>baseline (no regularization)</td><td class="num"></td><td class="num">0.380 (overfit)</td></tr>
        <tr><td>B</td><td>+ SpecAugment / noise / mixup</td><td class="num">0.297</td><td class="num">0.311 (underfit)</td></tr>
        <tr><td>C</td><td>fixed SpecAugment time-mask scale</td><td class="num">0.323</td><td class="num">0.343</td></tr>
        <tr><td>D</td><td>+ softened class weights (counts⁻⁰·⁵)</td><td class="num"><strong>0.330</strong></td><td class="num"><strong>0.376</strong></td></tr>
      </tbody>
    </table>
  </div>

  <h2>Known weak spots</h2>
  <ul class="weak">
    <li>Engines with many cylinders (8/12) are confused with close siblings far
      more than 1–2 cylinder engines, which have more distinctive, widely-spaced
      firing pulses.</li>
    <li>A couple of classes (<code>v12</code>, <code>2_rotor</code>) remain
      poorly calibrated even after two rounds of fixes.</li>
    <li>Cylinder-family accuracy (does it at least get the cylinder count
      right) is meaningfully higher than exact-class accuracy — most confusion
      is between siblings in the same family, not wild misfires.</li>
  </ul>

  <h2>Links</h2>
  <div class="links">
    <a href="https://huggingface.co/joakes90/engine_sound_cassifier" target="_blank" rel="noopener">Model repo (checkpoint + card)</a>
    <a href="https://github.com/joakes90/auto_sound_train" target="_blank" rel="noopener">Training code + notebook (GitHub)</a>
    <a href="https://github.com/joakes90/auto_sound_train/blob/master/FINDINGS.md" target="_blank" rel="noopener">Full diagnostic write-up</a>
  </div>

  <footer>
    CC BY-SA 4.0. This Space is static — no inference runs here.
  </footer>
</main>
</body>
</html>