Spaces:
Running
Running
| title: OmniAgentBench Voice Demo | |
| emoji: 🗣️ | |
| colorFrom: indigo | |
| colorTo: purple | |
| sdk: static | |
| pinned: false | |
| # OmniAgentBench — accent & voice demo | |
| Audio samples for feedback on speech variation in the benchmark. Two parts: | |
| - **`edge/`** — 14 real English accents (American, British, Australian, Canadian, | |
| Irish, Indian, South African, Nigerian, Kenyan, New Zealand, Filipino, | |
| Singaporean, Hong Kong, Tanzanian), synthesized with free Microsoft neural | |
| voices, on our actual benchmark instructions. | |
| - **`tts/`** — Qwen3-TTS VoiceDesign controls: speaking rate, gender, age, pitch, | |
| emotion, and 41 accent *attempts* (accent does not render — every one sounds the | |
| same neutral voice, which is why we moved to per-accent voices for real accents). | |
| Open the Space to listen. We are gathering opinions on which accents sound natural | |
| and usable to add as a speech-variation dimension. | |
| Audio here is synthesized (Microsoft / Qwen neural voices) for demonstration only. | |