Repository navigation
Expand file tree
/
Copy pathrobot.html
More file actions
176 lines (162 loc) · 8.45 KB
/
Copy pathrobot.html
File metadata and controls
176 lines (162 loc) · 8.45 KB
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>DUET: Representation Steering for Emotion Control in Speech Synthesis</title>
<style>
:root {
--accent: #2563eb;
--text: #1e293b;
--text-sec: #64748b;
--bg: #ffffff;
--bg-alt: #f8fafc;
--border: #e2e8f0;
--angry: #dc2626; --angry-bg: #fef2f2;
--happy: #ca8a04; --happy-bg: #fefce8;
--sad: #2563eb; --sad-bg: #eff6ff;
--radius: 8px;
}
* { box-sizing: border-box; margin: 0; padding: 0; }
body { font-family: 'Segoe UI', system-ui, -apple-system, sans-serif;
background: var(--bg-alt); color: var(--text); line-height: 1.7;
-webkit-font-smoothing: antialiased; }
/* ── Hero ── */
.hero { background: linear-gradient(135deg, #0f172a 0%, #1e3a5f 100%);
color: #fff; padding: 56px 24px 48px; text-align: center; }
.hero .method { font-size: 1.1em; color: #93c5fd; font-weight: 600; margin-bottom: 6px; }
.hero h1 { font-size: 2em; font-weight: 700; letter-spacing: -0.02em;
max-width: 720px; margin: 0 auto 6px; }
.hero .authors { font-size: 0.92em; color: #94a3b8; margin-bottom: 20px; }
.hero .links { display: flex; gap: 12px; justify-content: center; flex-wrap: wrap; }
.hero .links a { display: inline-flex; align-items: center; gap: 6px;
padding: 8px 18px; border-radius: 20px; font-size: 0.88em;
font-weight: 600; text-decoration: none; transition: all 0.2s;
background: rgba(255,255,255,0.12); color: #fff;
border: 1px solid rgba(255,255,255,0.22); }
.hero .links a:hover { background: rgba(255,255,255,0.22); }
.hero .links a.primary { background: var(--accent); border-color: var(--accent); }
.hero .links a.primary:hover { background: #1d4ed8; }
/* ── Sections ── */
section { background: var(--bg); border-radius: 12px; padding: 36px;
box-shadow: 0 1px 3px rgba(0,0,0,0.04), 0 4px 16px rgba(0,0,0,0.03);
margin: 20px auto; max-width: 1100px; }
section h2 { font-size: 1.3em; font-weight: 700; margin-bottom: 16px;
display: flex; align-items: center; gap: 10px; }
section h2::before { content: ''; width: 4px; height: 24px; background: var(--accent);
border-radius: 2px; flex-shrink: 0; }
/* ── Abstract ── */
.abstract { border-left: 3px solid var(--accent); padding-left: 20px;
color: var(--text-sec); font-size: 0.93em; line-height: 1.8; }
/* ── Video ── */
.emo-section { margin: 28px 0 0; }
.emo-label { display: inline-block; padding: 4px 14px; border-radius: 16px;
font-size: 0.88em; font-weight: 700; margin-bottom: 12px; }
.emo-label-angry { background: var(--angry-bg); color: var(--angry); }
.emo-label-happy { background: var(--happy-bg); color: var(--happy); }
.emo-label-sad { background: var(--sad-bg); color: var(--sad); }
.video-grid { display: grid; grid-template-columns: repeat(4, 1fr); gap: 14px; }
.video-card { background: var(--bg-alt); border-radius: var(--radius); overflow: hidden;
border: 1px solid var(--border); transition: box-shadow 0.2s; }
.video-card:hover { box-shadow: 0 4px 16px rgba(0,0,0,0.08); }
.video-card video { width: 100%; display: block; }
.video-card .meta { padding: 8px 14px; font-size: 0.82em; color: var(--text-sec); }
/* ── Footer ── */
footer { text-align: center; padding: 36px 20px; color: var(--text-sec); font-size: 0.82em; }
@media (max-width: 900px) {
.video-grid { grid-template-columns: repeat(2, 1fr); }
}
@media (max-width: 640px) {
.hero h1 { font-size: 1.4em; }
section { padding: 24px; margin: 12px 16px; }
.video-grid { grid-template-columns: 1fr; }
}
</style>
</head>
<body>
<div class="hero">
<div class="method">DUET</div>
<h1>Representation Steering for Emotion Control in Speech Synthesis</h1>
<p class="authors">Anonymous Authors — NeurIPS 2026 Submission</p>
<div class="links">
<a href="#">Paper (PDF)</a>
<a href="#robot-demo" class="primary">Robot Demo</a>
</div>
</div>
<section id="abstract">
<h2>Abstract</h2>
<p class="abstract">
Text-to-speech (TTS) models have achieved human level naturalness, particularly in diffusion and flow-matching models whose iterative denoising captures the high dimensional, richly detailed distribution of speech. However, existing methods for emotion control still rely on supervised retraining on emotional speech corpora with fixed emotion classes, because emotional cues are subtle, correlate with speaker identity, and diverge across architectures. We discover that emotion is linearly decodable from pretrained TTS hidden states and nearly orthogonal to speaker identity, despite no emotion supervision. Building on this, we introduce DUET, a plug-and-play dual space framework for emotion control in frozen TTS models. DUET applies two complementary interventions at each denoising step, (i) steering the hidden state along a target emotion direction obtained by probing for the layer most discriminative of emotion, and (ii) guiding the clean estimate via a gradient backpropagated through a differentiable vocoder to refine spectral details that hidden state steering alone cannot reach. We evaluate DUET on 3 datasets across 5 TTS architectures, with quantitative results showing clear advantages over 10 baselines and the highest emotion appropriateness ratings in human evaluation. To further showcase its qualitative behavior, we deploy DUET on an Ameca humanoid robot, where it produces richly expressive emotional speech on real hardware, demonstrating the strong potential for plug-and-play affective interaction for embodied agents.
</p>
</section>
<section id="robot-demo">
<h2>Robot Deployment</h2>
<div class="emo-section">
<span class="emo-label emo-label-angry">Angry</span>
<div class="video-grid">
<div class="video-card">
<video controls preload="metadata" src="video/angry_1.mp4"></video>
<div class="meta">Clip 1</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/angry_2.mp4"></video>
<div class="meta">Clip 2</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/angry_3.mp4"></video>
<div class="meta">Clip 3</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/angry_4.mp4"></video>
<div class="meta">Clip 4</div>
</div>
</div>
</div>
<div class="emo-section">
<span class="emo-label emo-label-happy">Happy</span>
<div class="video-grid">
<div class="video-card">
<video controls preload="metadata" src="video/happy_1.mp4"></video>
<div class="meta">Clip 1</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/happy_2.mp4"></video>
<div class="meta">Clip 2</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/happy_3.mp4"></video>
<div class="meta">Clip 3</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/happy_4.mp4"></video>
<div class="meta">Clip 4</div>
</div>
</div>
</div>
<div class="emo-section">
<span class="emo-label emo-label-sad">Sad</span>
<div class="video-grid">
<div class="video-card">
<video controls preload="metadata" src="video/sad_1.mp4"></video>
<div class="meta">Clip 1</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/sad_2.mp4"></video>
<div class="meta">Clip 2</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/sad_3.mp4"></video>
<div class="meta">Clip 3</div>
</div>
<div class="video-card">
<video controls preload="metadata" src="video/sad_4.mp4"></video>
<div class="meta">Clip 4</div>
</div>
</div>
</div>
</section>
<footer>
Anonymous submission to NeurIPS 2026 — Under double-blind review
</footer>
</body>
</html>