demo / index.html
Hodfa71's picture
Multi-turn: use actual released audio (v1 concat vs v2 per-turn)
3ab8ce0 verified
Raw
History Blame Contribute Delete
3.2 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>OmniAgentBench — accent & voice demo</title>
<style>
:root { color-scheme: light dark; }
body { font-family: system-ui,-apple-system,Segoe UI,Roboto,sans-serif; max-width: 720px;
margin: 2.5rem auto; padding: 0 1.2rem; line-height: 1.55; }
h1 { font-size: 1.6rem; margin-bottom: .3rem; }
.lead { opacity: .8; margin-bottom: 2rem; }
.card { display: block; border: 1px solid rgba(127,127,127,.3); border-radius: 12px;
padding: 1.1rem 1.3rem; margin: 1rem 0; text-decoration: none; color: inherit; }
.card:hover { border-color: rgba(127,127,127,.7); }
.card h2 { margin: 0 0 .3rem; font-size: 1.15rem; }
.card p { margin: 0; opacity: .78; font-size: .92rem; }
.tag { display: inline-block; font-size: .72rem; padding: .1rem .5rem; border-radius: 999px;
background: rgba(127,127,127,.18); margin-left: .4rem; vertical-align: middle; }
.ask { margin-top: 2rem; padding: .9rem 1.1rem; border-left: 3px solid #8a7cff;
background: rgba(138,124,255,.08); border-radius: 6px; font-size: .92rem; }
</style>
</head>
<body>
<h1>OmniAgentBench — accent &amp; voice demo 🗣️</h1>
<p class="lead">Speech-variation samples on our actual benchmark instructions.
Listen and tell us what sounds natural and usable.</p>
<a class="card" href="edge/index.html">
<h2>Real accents <span class="tag">14 accents · free</span></h2>
<p>Genuine per-accent English voices (American, British, Indian, Nigerian,
Singaporean, and 9 more) via free Microsoft neural voices. These are the accents
we would actually add to the benchmark.</p>
</a>
<a class="card" href="tts/index.html">
<h2>Qwen3-TTS voice controls <span class="tag">rate · gender · age · pitch · emotion</span></h2>
<p>What the instruct-driven TTS can control — speaking rate, gender, age, pitch,
and emotion all work. It also shows 41 accent <em>attempts</em>: notice they all
sound like one neutral voice. That is why real accents come from per-accent voices
instead.</p>
</a>
<a class="card" href="multi/index.html">
<h2>Multi-turn: v1 vs v2 <span class="tag">actual released audio</span></h2>
<p>Straight from the dataset: v1 is the single concatenated clip the evaluation
used, v2 is the same instruction as separate per-turn clips (a real conversation).
Both files already ship in the release, so the difference is purely how the turns
are delivered to the model.</p>
</a>
<a class="card" href="noise/index.html">
<h2>Noise: actual vs new <span class="tag">fix comparison</span></h2>
<p>The actual released noise (downloaded from the dataset) measures about 8 dB and
masks the speech. The fix measures the active region and targets 15 dB, so noise
stays in the background. Actual vs new per environment, same clip, with measured SNR.</p>
</a>
<div class="ask">
<strong>Feedback wanted:</strong> which accents sound natural and clear enough to
include as a speech-variation dimension? Any that sound wrong or unintelligible?
And for multi-turn, does v1 or v2 feel like the fairer test?
</div>
</body>
</html>