alexshah commited on
Commit
72cc41d
·
1 Parent(s): 14d1ab6

fix colors

Browse files
Files changed (2) hide show
  1. index.html +250 -249
  2. style.css +16 -0
index.html CHANGED
@@ -1,264 +1,265 @@
1
  <!doctype html>
2
  <html lang="en">
3
- <head>
4
- <meta charset="utf-8" />
5
- <meta name="viewport" content="width=device-width, initial-scale=1" />
6
- <meta
7
- name="description"
8
- content="ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks."
9
- />
10
- <title>ArmBench-ASR</title>
11
- <script>
12
- (() => {
13
- let saved = null;
14
- try { saved = localStorage.getItem("arm-asr-theme"); } catch (_) {}
15
- const theme = saved || (matchMedia("(prefers-color-scheme: dark)").matches ? "dark" : "light");
16
- document.documentElement.dataset.theme = theme;
17
- })();
18
- </script>
19
- <link rel="preconnect" href="https://fonts.googleapis.com" />
20
- <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
21
- <link
22
- href="https://fonts.googleapis.com/css2?family=DM+Sans:wght@400;500;600;700&amp;family=Manrope:wght@600;700;800&amp;display=swap"
23
- rel="stylesheet"
24
- />
25
- <link rel="stylesheet" href="style.css" />
26
- <script src="app.js" defer></script>
27
- </head>
28
- <body>
29
- <main class="shell">
30
- <header class="hero">
31
- <div class="hero-tools">
32
- <button class="theme-toggle" id="theme-toggle" type="button" aria-label="Switch color theme">
33
- <svg class="sun-icon" viewBox="0 0 24 24" aria-hidden="true">
34
- <circle cx="12" cy="12" r="3.5"></circle>
35
- <path d="M12 2v2m0 16v2M4.9 4.9l1.4 1.4m11.4 11.4 1.4 1.4M2 12h2m16 0h2M4.9 19.1l1.4-1.4M17.7 6.3l1.4-1.4"></path>
36
- </svg>
37
- <svg class="moon-icon" viewBox="0 0 24 24" aria-hidden="true">
38
- <path d="M20.2 15.1A8.5 8.5 0 0 1 8.9 3.8a8.5 8.5 0 1 0 11.3 11.3Z"></path>
39
- </svg>
40
- </button>
41
- </div>
42
- <div class="hero-copy">
43
- <h1><span class="heading-flag" role="img" aria-label="Armenian flag">🇦🇲</span> ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks</h1>
44
- <p class="lede">
45
- <span class="version-badge">v0.1</span>
46
- </p>
47
- </div>
48
- <div class="hero-art" aria-hidden="true">
49
- <span></span><span></span><span></span><span></span><span></span>
50
- <span></span><span></span><span></span><span></span>
51
- </div>
52
- </header>
53
-
54
- <nav class="tabs" role="tablist" aria-label="Benchmark sections">
55
- <button class="tab is-active" role="tab" aria-selected="true" data-tab="leaderboards">
56
- Leaderboards
57
- </button>
58
- <button class="tab" role="tab" aria-selected="false" data-tab="datasets">Datasets</button>
59
- <button class="tab" role="tab" aria-selected="false" data-tab="about">About</button>
60
- </nav>
61
-
62
- <section class="panel is-active matrix-view" id="leaderboards" role="tabpanel">
63
- <div class="table-heading">
64
- <div>
65
- <h2 id="table-title">Overall leaderboard</h2>
66
- <p id="table-description">Rank is based on WER. Scores are percentages; lower is better.</p>
67
- </div>
68
- <div class="table-tools">
69
- <button class="color-toggle" id="color-toggle" type="button" aria-pressed="false">
70
- <span class="color-swatch good"></span><span class="color-swatch mid"></span><span class="color-swatch bad"></span>
71
- Colors
72
- </button>
73
- <span class="result-count" id="result-count" aria-live="polite"></span>
74
- </div>
75
- </div>
76
-
77
- <div class="table-controls" aria-label="Leaderboard display and filters">
78
- <div class="view-switch" role="group" aria-label="Leaderboard view">
79
- <button class="view-button is-active" type="button" data-view="matrix" aria-pressed="true">All datasets</button>
80
- <button class="view-button" type="button" data-view="single" aria-pressed="false">Single dataset</button>
81
- </div>
82
-
83
- <div class="filter-card">
84
- <fieldset class="field dataset-field">
85
- <legend>Dataset</legend>
86
- <div class="choice-options" id="dataset-options"></div>
87
- </fieldset>
88
-
89
- <fieldset class="field matrix-metric-field">
90
- <legend>Metric across all datasets</legend>
91
- <div class="choice-options" id="matrix-metric-options"></div>
92
- </fieldset>
93
- </div>
94
- </div>
95
-
96
- <div class="table-card">
97
- <div class="table-scroll">
98
- <table id="leaderboard-table">
99
- <thead></thead>
100
- <tbody></tbody>
101
- </table>
102
- </div>
103
- <div class="empty-state" id="empty-state" hidden>
104
- <h3>No matching models</h3>
105
- <p>Select Open or Closed in the Availability column to show models.</p>
106
- </div>
107
- </div>
108
-
109
- <section class="comparison-chart" aria-labelledby="comparison-chart-title">
110
- <div class="comparison-chart-heading">
111
- <div>
112
- <h2 id="comparison-chart-title">Model comparison</h2>
113
- <p>Compare model error rates for one dataset and metric. Lower is better.</p>
114
- </div>
115
- <div class="comparison-chart-controls" aria-label="Model comparison filters">
116
- <label>
117
- <span>Dataset</span>
118
- <select id="chart-dataset" aria-label="Chart dataset"></select>
119
- </label>
120
- <label>
121
- <span>Metric</span>
122
- <select id="chart-metric" aria-label="Chart metric"></select>
123
- </label>
124
- <label>
125
- <span>Availability</span>
126
- <select id="chart-availability" aria-label="Chart model availability">
127
- <option value="All">All models</option>
128
- <option value="Open">Open only</option>
129
- <option value="Closed">Closed only</option>
130
- </select>
131
- </label>
132
- </div>
133
- </div>
134
- <div class="comparison-chart-card">
135
- <div class="comparison-bars" id="comparison-bars" aria-live="polite"></div>
136
- <p class="comparison-chart-empty" id="comparison-chart-empty" hidden>No models match the current filters.</p>
137
- </div>
138
- </section>
139
-
140
- <section class="citation-card" aria-labelledby="citation-title">
141
- <p class="citation-label" id="citation-title">Cite ArmBench-ASR</p>
142
- <p>If you use ArmBench-ASR in your research, please cite it as follows:</p>
143
- <pre><code>@misc{armbench-asr,
144
  title={ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks},
145
  author={Metric-AI-Lab},
146
  year={2026},
147
  howpublished={\url{https://huggingface.co/spaces/Metric-AI/ArmBench-ASR}},
148
  note={Benchmark for Evaluating Speech-to-Text Models on Armenian Speech Recognition Tasks}
149
  }</code></pre>
150
- </section>
151
- </section>
152
-
153
- <section class="panel" id="datasets" role="tabpanel" hidden>
154
- <div class="section-intro">
155
- <h2>Five datasets, one Armenian benchmark</h2>
156
- <p>10,113 clips and 20.7 hours of speech spanning read, literary, cinematic, and conversational Armenian. Every model is evaluated on the same fixed test material.</p>
157
- </div>
158
- <div class="dataset-grid">
159
- <article class="dataset-card">
160
- <div class="dataset-card-top"><span>01</span><span class="dataset-tag public">Public</span></div>
161
- <h3><a class="dataset-source" href="https://mozilladatacollective.com/datasets/cmqinsp2y00xsnr07ra8gxfoi" target="_blank" rel="noreferrer">Common Voice 26 <span>↗</span></a></h3>
162
- <p>Mozilla Common Voice Scripted Speech 26.0 for Armenian, using the full <code>hy-AM</code> test split released in June 2026.</p>
163
- <dl class="dataset-meta"><div><dt>Samples</dt><dd>6,892</dd></div><div><dt>Audio</dt><dd>10.8 h</dd></div><div><dt>Style</dt><dd>Read speech</dd></div></dl>
164
- </article>
165
-
166
- <article class="dataset-card">
167
- <div class="dataset-card-top"><span>02</span><span class="dataset-tag public">Public</span></div>
168
- <h3><a class="dataset-source" href="https://huggingface.co/datasets/google/fleurs" target="_blank" rel="noreferrer">FLEURS <span>↗</span></a></h3>
169
- <p>Google FLEURS Armenian (<code>hy_am</code>) test split. Raw transcriptions retain punctuation and capitalization for strict scoring, with some <a href="https://huggingface.co/datasets/Metric-AI/fleurs-corrections" target="_blank" rel="noreferrer">manual corrections</a> applied.</p>
170
- <dl class="dataset-meta"><div><dt>Samples</dt><dd>932</dd></div><div><dt>Audio</dt><dd>3.0 h</dd></div><div><dt>Style</dt><dd>Read speech</dd></div></dl>
171
- </article>
172
-
173
- <article class="dataset-card">
174
- <div class="dataset-card-top"><span>03</span><span class="dataset-tag private">Private</span></div>
175
- <h3>Poems</h3>
176
- <p>Read Armenian poetry with background music, literary vocabulary, expressive phrasing, and text structures underrepresented in general ASR sets.</p>
177
- <dl class="dataset-meta"><div><dt>Samples</dt><dd>969</dd></div><div><dt>Audio</dt><dd>3.3 h</dd></div><div><dt>Style</dt><dd>Literary</dd></div></dl>
178
- </article>
179
-
180
- <article class="dataset-card">
181
- <div class="dataset-card-top"><span>04</span><span class="dataset-tag private">Private</span></div>
182
- <h3>Movies</h3>
183
- <p>Armenian movie dialogue with conversational delivery, produced-media acoustics, and occasional background noise.</p>
184
- <dl class="dataset-meta"><div><dt>Samples</dt><dd>820</dd></div><div><dt>Audio</dt><dd>1.7 h</dd></div><div><dt>Style</dt><dd>Dialogue</dd></div></dl>
185
- </article>
186
-
187
- <article class="dataset-card">
188
- <div class="dataset-card-top"><span>05</span><span class="dataset-tag private">Private</span></div>
189
- <h3>News</h3>
190
- <p>A compact Armenian news dataset recited by a speaker, designed to evaluate clear narrated speech beyond standard public test sets.</p>
191
- <dl class="dataset-meta"><div><dt>Samples</dt><dd>500</dd></div><div><dt>Audio</dt><dd>1.9 h</dd></div><div><dt>Style</dt><dd>Narrated news</dd></div></dl>
192
- </article>
193
- </div>
194
-
195
- <section class="citation-card" aria-labelledby="citation-title">
196
- <p class="citation-label" id="citation-title">Cite ArmBench-ASR</p>
197
- <p>If you use ArmBench-ASR in your research, please cite it as follows:</p>
198
- <pre><code>@misc{armbench-asr,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
199
  title={ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks},
200
  author={Metric-AI-Lab},
201
  year={2026},
202
  howpublished={\url{https://huggingface.co/spaces/Metric-AI/ArmBench-ASR}},
203
  note={Benchmark for Evaluating Speech-to-Text Models on Armenian Speech Recognition Tasks}
204
  }</code></pre>
205
- </section>
206
- </section>
207
-
208
- <section class="panel" id="about" role="tabpanel" hidden>
209
- <div class="about-grid">
210
- <div class="section-intro">
211
- <h2>Reading the benchmark</h2>
212
- <p>
213
- Armenian ASR Benchmark compares automatic speech recognition systems on a varied
214
- Armenian evaluation suite. Every score is an error rate, so lower values are better.
215
- </p>
216
- <p>
217
- The benchmark aims to bring clarity to the Armenian ASR landscape, make model comparisons easier, and help model creators identify strengths, gaps, and opportunities for meaningful improvement.
218
- </p>
219
- </div>
220
- <div class="definition-list">
221
- <article><strong>WER</strong><span>Word error rate after strict, punctuation-preserving preprocessing.</span></article>
222
- <article><strong>CER</strong><span>Character error rate after strict, punctuation-preserving preprocessing.</span></article>
223
- <article><strong>WER-norm</strong><span>Word error rate after text normalization.</span></article>
224
- <article><strong>CER-norm</strong><span>Character error rate after text normalization.</span></article>
225
- </div>
226
- </div>
227
- <div class="note">
228
- <strong>Availability</strong>
229
- <p><b>Open</b> means model weights are publicly available. <b>Closed</b> means access is provided through a hosted product or API.</p>
230
- </div>
231
- <div class="note">
232
- <strong>API settings</strong>
233
- <p>Hosted systems use fixed model identifiers and transcription-only prompts. Evaluated Gemini runs use temperature 0, a 1,000-token output cap, and audio limited to 40 seconds. Thinking is disabled or set to the lowest applicable level: budget 0 for Gemini 2.5 Flash, budget 128 for Gemini 2.5 Pro, and <b>MINIMAL</b> for Gemini 3.1 Flash Lite, 3.5 Flash Lite, and 3.6 Flash.</p>
234
- </div>
235
- <div class="note">
236
- <strong>Text cleaning</strong>
237
- <p>References and predictions are processed with the same deterministic pipeline. Strict scoring standardizes punctuation, spacing, capitalization, and Unicode; normalized scoring additionally reduces orthographic and formatting differences while preserving Armenian, English, Russian letters, digits, and recognized clock times. Both <b>WER / CER</b> and <b>WER-norm / CER-norm</b> are reported.</p>
238
- </div>
239
-
240
- <section class="lab-card">
241
- <div>
242
- <h2>About Us</h2>
243
- <p>Armenian ASR Benchmark is a Metric AI Lab project. The applied AI research lab builds production systems across custom language and vision-language models, multilingual voice agents, Physical AI, retrieval, and analytics—while publishing frontier research and open benchmarks.</p>
244
- </div>
245
- <a href="https://metricailab.com" target="_blank" rel="noreferrer">Visit Metric AI Lab <span>↗</span></a>
246
- </section>
247
-
248
- <section class="acknowledgments">
249
- <p class="acknowledgments-label">Acknowledgments</p>
250
- <p>
251
- We gratefully acknowledge <strong>HiSpeech</strong>, <strong>Talk2Edit</strong>,
252
- <strong>TBB ASR</strong>, and <strong>Wav.am</strong> for making their systems
253
- available for evaluation at no cost, and <strong>Infocom</strong> for providing
254
- Armenian news data used in this benchmark.
255
- </p>
256
- </section>
257
-
258
- <footer>
259
- <span>Armenian ASR Benchmark · v0.1</span>
260
- <span>A Metric AI Lab project · Built for transparent Armenian speech recognition evaluation.</span>
261
- </footer>
262
- </main>
263
- </body>
264
  </html>
 
1
  <!doctype html>
2
  <html lang="en">
3
+ <head>
4
+ <meta charset="utf-8" />
5
+ <meta name="viewport" content="width=device-width, initial-scale=1" />
6
+ <meta name="description" content="ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks." />
7
+ <title>ArmBench-ASR</title>
8
+ <script>
9
+ (() => {
10
+ let saved = null;
11
+ try {
12
+ saved = localStorage.getItem("arm-asr-theme");
13
+ } catch (_) {}
14
+ const theme = saved || (matchMedia("(prefers-color-scheme: dark)").matches ? "dark" : "light");
15
+ document.documentElement.dataset.theme = theme;
16
+ })();
17
+ </script>
18
+ <link rel="preconnect" href="https://fonts.googleapis.com" />
19
+ <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
20
+ <link href="https://fonts.googleapis.com/css2?family=DM+Sans:wght@400;500;600;700&amp;family=Manrope:wght@600;700;800&amp;display=swap" rel="stylesheet" />
21
+ <link rel="stylesheet" href="style.css" />
22
+ <script src="app.js" defer></script>
23
+ </head>
24
+ <body>
25
+ <main class="shell">
26
+ <header class="hero">
27
+ <div class="hero-tools"><button class="theme-toggle" id="theme-toggle" type="button" aria-label="Switch color theme"><svg class="sun-icon" viewBox="0 0 24 24" aria-hidden="true">
28
+ <circle cx="12" cy="12" r="3.5"></circle>
29
+ <path d="M12 2v2m0 16v2M4.9 4.9l1.4 1.4m11.4 11.4 1.4 1.4M2 12h2m16 0h2M4.9 19.1l1.4-1.4M17.7 6.3l1.4-1.4"></path>
30
+ </svg><svg class="moon-icon" viewBox="0 0 24 24" aria-hidden="true">
31
+ <path d="M20.2 15.1A8.5 8.5 0 0 1 8.9 3.8a8.5 8.5 0 1 0 11.3 11.3Z"></path>
32
+ </svg></button></div>
33
+ <div class="hero-copy">
34
+ <h1><span class="heading-flag" role="img" aria-label="Armenian flag">🇦🇲</span> ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks</h1>
35
+ <p class="lede"><span class="version-badge">v0.1</span></p>
36
+ </div>
37
+ <div class="hero-art" aria-hidden="true"><span></span><span></span><span></span><span></span><span></span><span></span><span></span><span></span><span></span></div>
38
+ </header>
39
+ <nav class="tabs" role="tablist" aria-label="Benchmark sections"><button class="tab is-active" role="tab" aria-selected="true" data-tab="leaderboards"> Leaderboards </button><button class="tab" role="tab" aria-selected="false" data-tab="datasets">Datasets</button><button class="tab" role="tab" aria-selected="false" data-tab="about">About</button></nav>
40
+ <section class="panel is-active matrix-view" id="leaderboards" role="tabpanel">
41
+ <div class="table-heading">
42
+ <div>
43
+ <h2 id="table-title">Overall leaderboard</h2>
44
+ <p id="table-description"> Rank is based on WER. Scores are percentages; lower is better. </p>
45
+ </div>
46
+ <div class="table-tools"><span class="result-count" id="result-count" aria-live="polite"></span></div>
47
+ </div>
48
+ <div class="table-controls" aria-label="Leaderboard display and filters">
49
+ <div class="view-switch" role="group" aria-label="Leaderboard view"><button class="view-button is-active" type="button" data-view="matrix" aria-pressed="true"> All datasets </button><button class="view-button" type="button" data-view="single" aria-pressed="false"> Single dataset </button></div>
50
+ <div class="filter-card">
51
+ <fieldset class="field dataset-field">
52
+ <legend>Dataset</legend>
53
+ <div class="choice-options" id="dataset-options"></div>
54
+ </fieldset>
55
+ <fieldset class="field matrix-metric-field">
56
+ <legend>Metric across all datasets</legend>
57
+ <div class="choice-options" id="matrix-metric-options"></div>
58
+ </fieldset>
59
+ <div class="filter-actions"><button class="color-toggle" id="color-toggle" type="button" aria-pressed="false"><span class="color-swatch good"></span><span class="color-swatch mid"></span><span class="color-swatch bad"></span> Colors </button></div>
60
+ </div>
61
+ </div>
62
+ <div class="table-card">
63
+ <div class="table-scroll">
64
+ <table id="leaderboard-table">
65
+ <thead></thead>
66
+ <tbody></tbody>
67
+ </table>
68
+ </div>
69
+ <div class="empty-state" id="empty-state" hidden>
70
+ <h3>No matching models</h3>
71
+ <p>Select Open or Closed in the Availability column to show models.</p>
72
+ </div>
73
+ </div>
74
+ <div class="table-card">
75
+ <div class="table-scroll">
76
+ <table id="leaderboard-table">
77
+ <thead></thead>
78
+ <tbody></tbody>
79
+ </table>
80
+ </div>
81
+ <div class="empty-state" id="empty-state" hidden>
82
+ <h3>No matching models</h3>
83
+ <p>Select Open or Closed in the Availability column to show models.</p>
84
+ </div>
85
+ </div>
86
+ <section class="comparison-chart" aria-labelledby="comparison-chart-title">
87
+ <div class="comparison-chart-heading">
88
+ <div>
89
+ <h2 id="comparison-chart-title">Model comparison</h2>
90
+ <p>Compare model error rates for one dataset and metric. Lower is better.</p>
91
+ </div>
92
+ <div class="comparison-chart-controls" aria-label="Model comparison filters"><label><span>Dataset</span><select id="chart-dataset" aria-label="Chart dataset"></select></label><label><span>Metric</span><select id="chart-metric" aria-label="Chart metric"></select></label><label><span>Availability</span><select id="chart-availability" aria-label="Chart model availability">
93
+ <option value="All">All models</option>
94
+ <option value="Open">Open only</option>
95
+ <option value="Closed">Closed only</option>
96
+ </select></label></div>
97
+ </div>
98
+ <div class="comparison-chart-card">
99
+ <div class="comparison-bars" id="comparison-bars" aria-live="polite"></div>
100
+ <p class="comparison-chart-empty" id="comparison-chart-empty" hidden>No models match the current filters.</p>
101
+ </div>
102
+ </section>
103
+ <section class="citation-card" aria-labelledby="citation-title">
104
+ <p class="citation-label" id="citation-title">Cite ArmBench-ASR</p>
105
+ <p>If you use ArmBench-ASR in your research, please cite it as follows:</p>
106
+ <pre><code>@misc{armbench-asr,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
107
  title={ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks},
108
  author={Metric-AI-Lab},
109
  year={2026},
110
  howpublished={\url{https://huggingface.co/spaces/Metric-AI/ArmBench-ASR}},
111
  note={Benchmark for Evaluating Speech-to-Text Models on Armenian Speech Recognition Tasks}
112
  }</code></pre>
113
+ </section>
114
+ </section>
115
+ <section class="panel" id="datasets" role="tabpanel" hidden>
116
+ <div class="section-intro">
117
+ <h2>Five datasets, one Armenian benchmark</h2>
118
+ <p>10,113 clips and 20.7 hours of speech spanning read, literary, cinematic, and conversational Armenian. Every model is evaluated on the same fixed test material.</p>
119
+ </div>
120
+ <div class="dataset-grid">
121
+ <article class="dataset-card">
122
+ <div class="dataset-card-top"><span>01</span><span class="dataset-tag public">Public</span></div>
123
+ <h3><a class="dataset-source" href="https://mozilladatacollective.com/datasets/cmqinsp2y00xsnr07ra8gxfoi" target="_blank" rel="noreferrer">Common Voice 26 <span>↗</span></a></h3>
124
+ <p>Mozilla Common Voice Scripted Speech 26.0 for Armenian, using the full <code>hy-AM</code> test split released in June 2026.</p>
125
+ <dl class="dataset-meta">
126
+ <div>
127
+ <dt>Samples</dt>
128
+ <dd>6,892</dd>
129
+ </div>
130
+ <div>
131
+ <dt>Audio</dt>
132
+ <dd>10.8 h</dd>
133
+ </div>
134
+ <div>
135
+ <dt>Style</dt>
136
+ <dd>Read speech</dd>
137
+ </div>
138
+ </dl>
139
+ </article>
140
+ <article class="dataset-card">
141
+ <div class="dataset-card-top"><span>02</span><span class="dataset-tag public">Public</span></div>
142
+ <h3><a class="dataset-source" href="https://huggingface.co/datasets/google/fleurs" target="_blank" rel="noreferrer">FLEURS <span>↗</span></a></h3>
143
+ <p>Google FLEURS Armenian (<code>hy_am</code>) test split. Raw transcriptions retain punctuation and capitalization for strict scoring, with some <a href="https://huggingface.co/datasets/Metric-AI/fleurs-corrections" target="_blank" rel="noreferrer">manual corrections</a> applied.</p>
144
+ <dl class="dataset-meta">
145
+ <div>
146
+ <dt>Samples</dt>
147
+ <dd>932</dd>
148
+ </div>
149
+ <div>
150
+ <dt>Audio</dt>
151
+ <dd>3.0 h</dd>
152
+ </div>
153
+ <div>
154
+ <dt>Style</dt>
155
+ <dd>Read speech</dd>
156
+ </div>
157
+ </dl>
158
+ </article>
159
+ <article class="dataset-card">
160
+ <div class="dataset-card-top"><span>03</span><span class="dataset-tag private">Private</span></div>
161
+ <h3>Poems</h3>
162
+ <p>Read Armenian poetry with background music, literary vocabulary, expressive phrasing, and text structures underrepresented in general ASR sets.</p>
163
+ <dl class="dataset-meta">
164
+ <div>
165
+ <dt>Samples</dt>
166
+ <dd>969</dd>
167
+ </div>
168
+ <div>
169
+ <dt>Audio</dt>
170
+ <dd>3.3 h</dd>
171
+ </div>
172
+ <div>
173
+ <dt>Style</dt>
174
+ <dd>Literary</dd>
175
+ </div>
176
+ </dl>
177
+ </article>
178
+ <article class="dataset-card">
179
+ <div class="dataset-card-top"><span>04</span><span class="dataset-tag private">Private</span></div>
180
+ <h3>Movies</h3>
181
+ <p>Armenian movie dialogue with conversational delivery, produced-media acoustics, and occasional background noise.</p>
182
+ <dl class="dataset-meta">
183
+ <div>
184
+ <dt>Samples</dt>
185
+ <dd>820</dd>
186
+ </div>
187
+ <div>
188
+ <dt>Audio</dt>
189
+ <dd>1.7 h</dd>
190
+ </div>
191
+ <div>
192
+ <dt>Style</dt>
193
+ <dd>Dialogue</dd>
194
+ </div>
195
+ </dl>
196
+ </article>
197
+ <article class="dataset-card">
198
+ <div class="dataset-card-top"><span>05</span><span class="dataset-tag private">Private</span></div>
199
+ <h3>News</h3>
200
+ <p>A compact Armenian news dataset recited by a speaker, designed to evaluate clear narrated speech beyond standard public test sets.</p>
201
+ <dl class="dataset-meta">
202
+ <div>
203
+ <dt>Samples</dt>
204
+ <dd>500</dd>
205
+ </div>
206
+ <div>
207
+ <dt>Audio</dt>
208
+ <dd>1.9 h</dd>
209
+ </div>
210
+ <div>
211
+ <dt>Style</dt>
212
+ <dd>Narrated news</dd>
213
+ </div>
214
+ </dl>
215
+ </article>
216
+ </div>
217
+ <section class="citation-card" aria-labelledby="citation-title">
218
+ <p class="citation-label" id="citation-title">Cite ArmBench-ASR</p>
219
+ <p>If you use ArmBench-ASR in your research, please cite it as follows:</p>
220
+ <pre><code>@misc{armbench-asr,
221
  title={ArmBench-ASR: Benchmarking Speech-to-Text Models on Armenian Speech Recognition Tasks},
222
  author={Metric-AI-Lab},
223
  year={2026},
224
  howpublished={\url{https://huggingface.co/spaces/Metric-AI/ArmBench-ASR}},
225
  note={Benchmark for Evaluating Speech-to-Text Models on Armenian Speech Recognition Tasks}
226
  }</code></pre>
227
+ </section>
228
+ </section>
229
+ <section class="panel" id="about" role="tabpanel" hidden>
230
+ <div class="about-grid">
231
+ <div class="section-intro">
232
+ <h2>Reading the benchmark</h2>
233
+ <p> Armenian ASR Benchmark compares automatic speech recognition systems on a varied Armenian evaluation suite. Every score is an error rate, so lower values are better. </p>
234
+ <p> The benchmark aims to bring clarity to the Armenian ASR landscape, make model comparisons easier, and help model creators identify strengths, gaps, and opportunities for meaningful improvement. </p>
235
+ </div>
236
+ <div class="definition-list">
237
+ <article><strong>WER</strong><span>Word error rate after strict, punctuation-preserving preprocessing.</span></article>
238
+ <article><strong>CER</strong><span>Character error rate after strict, punctuation-preserving preprocessing.</span></article>
239
+ <article><strong>WER-norm</strong><span>Word error rate after text normalization.</span></article>
240
+ <article><strong>CER-norm</strong><span>Character error rate after text normalization.</span></article>
241
+ </div>
242
+ </div>
243
+ <div class="note"><strong>Availability</strong>
244
+ <p><b>Open</b> means model weights are publicly available. <b>Closed</b> means access is provided through a hosted product or API.</p>
245
+ </div>
246
+ <div class="note"><strong>API settings</strong>
247
+ <p>Hosted systems use fixed model identifiers and transcription-only prompts. Evaluated Gemini runs use temperature 0, a 1,000-token output cap, and audio limited to 40 seconds. Thinking is disabled or set to the lowest applicable level: budget 0 for Gemini 2.5 Flash, budget 128 for Gemini 2.5 Pro, and <b>MINIMAL</b> for Gemini 3.1 Flash Lite, 3.5 Flash Lite, and 3.6 Flash.</p>
248
+ </div>
249
+ <div class="note"><strong>Text cleaning</strong>
250
+ <p>References and predictions are processed with the same deterministic pipeline. Strict scoring standardizes punctuation, spacing, capitalization, and Unicode; normalized scoring additionally reduces orthographic and formatting differences while preserving Armenian, English, Russian letters, digits, and recognized clock times. Both <b>WER / CER</b> and <b>WER-norm / CER-norm</b> are reported.</p>
251
+ </div>
252
+ <section class="lab-card">
253
+ <div>
254
+ <h2>About Us</h2>
255
+ <p>Armenian ASR Benchmark is a Metric AI Lab project. The applied AI research lab builds production systems across custom language and vision-language models, multilingual voice agents, Physical AI, retrieval, and analytics—while publishing frontier research and open benchmarks.</p>
256
+ </div><a href="https://metricailab.com" target="_blank" rel="noreferrer">Visit Metric AI Lab <span>↗</span></a>
257
+ </section>
258
+ <section class="acknowledgments">
259
+ <p class="acknowledgments-label">Acknowledgments</p>
260
+ <p> We gratefully acknowledge <strong>HiSpeech</strong>, <strong>Talk2Edit</strong>, <strong>TBB ASR</strong>, and <strong>Wav.am</strong> for making their systems available for evaluation at no cost, and <strong>Infocom</strong> for providing Armenian news data used in this benchmark. </p>
261
+ </section>
262
+ <footer><span>Armenian ASR Benchmark · v0.1</span><span>A Metric AI Lab project · Built for transparent Armenian speech recognition evaluation.</span></footer>
263
+ </main>
264
+ </body>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
265
  </html>
style.css CHANGED
@@ -1528,3 +1528,19 @@ html[data-theme="dark"] footer span:first-child {
1528
  html[data-theme="dark"] .comparison-availability-dot.closed {
1529
  background: #7d8795;
1530
  }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1528
  html[data-theme="dark"] .comparison-availability-dot.closed {
1529
  background: #7d8795;
1530
  }
1531
+ .filter-card {
1532
+ display: flex;
1533
+ align-items: flex-end;
1534
+ gap: 16px;
1535
+ flex-wrap: wrap;
1536
+ }
1537
+ .dataset-field,
1538
+ .matrix-metric-field {
1539
+ flex: 1 1 auto;
1540
+ width: auto;
1541
+ }
1542
+ .filter-actions {
1543
+ display: flex;
1544
+ align-items: center;
1545
+ margin-left: auto;
1546
+ }