| <svg xmlns="http://www.w3.org/2000/svg" width="1400" height="900" viewBox="0 0 1400 900"> | |
| <rect width="1400" height="900" rx="18" fill="#F8FBFF"/> | |
| <style>text{font-family:Inter,Segoe UI,Arial,sans-serif;fill:#10253F}.title{font-size:28px;font-weight:750;letter-spacing:-.5px}.subtitle{font-size:14px;fill:#61738A}.label{font-size:14px;font-weight:600}.small{font-size:12px;fill:#61738A}.value{font-size:14px;font-weight:750}.grid{stroke:#DCE6F1;stroke-width:1}</style> | |
| <rect x="0" y="0" width="8" height="900" rx="4" fill="#1769E0"/> | |
| <text class="title" x="56" y="52">Three-ASR intelligibility audit</text> | |
| <text class="subtitle" x="56" y="78">Semantic WER on the same 400 generated clips per system · lower is better</text> | |
| <circle cx="330.0" cy="112.0" r="6" fill="#1769E0"/> | |
| <text class="small" x="346" y="116">Qwen3-ASR</text> | |
| <rect x="534.0" y="106.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <text class="small" x="556" y="116">Nemotron 3.5</text> | |
| <path d="M 750.0 105.0 L 757.0 112.0 L 750.0 119.0 L 743.0 112.0 Z" fill="#0B2A54"/> | |
| <text class="small" x="766" y="116">Whisper large-v3</text> | |
| <line class="grid" x1="330.0" y1="160" x2="330.0" y2="805"/> | |
| <text class="small" x="330.0" y="152" text-anchor="middle">0%</text> | |
| <line class="grid" x1="457.0" y1="160" x2="457.0" y2="805"/> | |
| <text class="small" x="457.0" y="152" text-anchor="middle">2%</text> | |
| <line class="grid" x1="584.0" y1="160" x2="584.0" y2="805"/> | |
| <text class="small" x="584.0" y="152" text-anchor="middle">4%</text> | |
| <line class="grid" x1="711.0" y1="160" x2="711.0" y2="805"/> | |
| <text class="small" x="711.0" y="152" text-anchor="middle">6%</text> | |
| <line class="grid" x1="838.0" y1="160" x2="838.0" y2="805"/> | |
| <text class="small" x="838.0" y="152" text-anchor="middle">8%</text> | |
| <line class="grid" x1="965.0" y1="160" x2="965.0" y2="805"/> | |
| <text class="small" x="965.0" y="152" text-anchor="middle">10%</text> | |
| <text class="small" x="1040" y="152">QWEN</text> | |
| <text class="small" x="1125" y="152">NEMO</text> | |
| <text class="small" x="1210" y="152">WHISPER</text> | |
| <text class="label" x="56" y="210">Inflect-Micro-v2</text> | |
| <line x1="449.3" y1="188" x2="535.5" y2="188" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="490.1" cy="188.0" r="6" fill="#1769E0"/> | |
| <line x1="615.5" y1="205" x2="742.6" y2="205" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="670.2" y="199.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="459.9" y1="222" x2="551.7" y2="222" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 503.1 215.0 L 510.1 222.0 L 503.1 229.0 L 496.1 222.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="210">2.52%</text> | |
| <text class="value" x="1125" y="210">5.45%</text> | |
| <text class="value" x="1210" y="210">2.73%</text> | |
| <text class="label" x="56" y="286">Inflect-Nano-v2</text> | |
| <line x1="464.9" y1="264" x2="553.3" y2="264" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="506.8" cy="264.0" r="6" fill="#1769E0"/> | |
| <line x1="629.9" y1="281" x2="750.7" y2="281" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="681.4" y="275.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="458.7" y1="298" x2="542.6" y2="298" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 498.5 291.0 L 505.5 298.0 L 498.5 305.0 L 491.5 298.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="286">2.78%</text> | |
| <text class="value" x="1125" y="286">5.63%</text> | |
| <text class="value" x="1210" y="286">2.65%</text> | |
| <text class="label" x="56" y="362">KittenTTS Nano · Bruno</text> | |
| <line x1="428.0" y1="340" x2="510.9" y2="340" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="466.8" cy="340.0" r="6" fill="#1769E0"/> | |
| <line x1="532.6" y1="357" x2="635.2" y2="357" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="575.3" y="351.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="429.3" y1="374" x2="510.8" y2="374" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 467.7 367.0 L 474.7 374.0 L 467.7 381.0 L 460.7 374.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="362">2.15%</text> | |
| <text class="value" x="1125" y="362">3.96%</text> | |
| <text class="value" x="1210" y="362">2.17%</text> | |
| <text class="label" x="56" y="438">KittenTTS Nano · Hugo</text> | |
| <line x1="440.0" y1="416" x2="527.3" y2="416" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="481.7" cy="416.0" r="6" fill="#1769E0"/> | |
| <line x1="522.3" y1="433" x2="625.9" y2="433" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="565.0" y="427.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="426.0" y1="450" x2="506.5" y2="450" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 464.0 443.0 L 471.0 450.0 L 464.0 457.0 L 457.0 450.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="438">2.39%</text> | |
| <text class="value" x="1125" y="438">3.80%</text> | |
| <text class="value" x="1210" y="438">2.11%</text> | |
| <text class="label" x="56" y="514">Piper Low · Ryan</text> | |
| <line x1="465.8" y1="492" x2="557.0" y2="492" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="508.7" cy="492.0" r="6" fill="#1769E0"/> | |
| <line x1="625.3" y1="509" x2="739.3" y2="509" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="673.9" y="503.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="470.4" y1="526" x2="558.7" y2="526" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 512.4 519.0 L 519.4 526.0 L 512.4 533.0 L 505.4 526.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="514">2.81%</text> | |
| <text class="value" x="1125" y="514">5.51%</text> | |
| <text class="value" x="1210" y="514">2.87%</text> | |
| <text class="label" x="56" y="590">Piper Low · Danny</text> | |
| <line x1="456.9" y1="568" x2="542.1" y2="568" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="496.6" cy="568.0" r="6" fill="#1769E0"/> | |
| <line x1="629.9" y1="585" x2="744.1" y2="585" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="679.5" y="579.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="452.0" y1="602" x2="537.0" y2="602" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 491.9 595.0 L 498.9 602.0 L 491.9 609.0 L 484.9 602.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="590">2.62%</text> | |
| <text class="value" x="1125" y="590">5.60%</text> | |
| <text class="value" x="1210" y="590">2.55%</text> | |
| <text class="label" x="56" y="666">Supertonic 3 · M2 · 3-step</text> | |
| <line x1="480.9" y1="644" x2="569.2" y2="644" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="522.6" cy="644.0" r="6" fill="#1769E0"/> | |
| <line x1="655.7" y1="661" x2="776.1" y2="661" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="707.4" y="655.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="488.3" y1="678" x2="586.5" y2="678" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 534.7 671.0 L 541.7 678.0 L 534.7 685.0 L 527.7 678.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="666">3.03%</text> | |
| <text class="value" x="1125" y="666">6.04%</text> | |
| <text class="value" x="1210" y="666">3.22%</text> | |
| <text class="label" x="56" y="742">Supertonic 3 · M2 · 8-step</text> | |
| <line x1="423.2" y1="720" x2="503.7" y2="720" stroke="#1769E0" stroke-width="2"/> | |
| <circle cx="460.3" cy="720.0" r="6" fill="#1769E0"/> | |
| <line x1="511.4" y1="737" x2="605.7" y2="737" stroke="#6EA8F2" stroke-width="2"/> | |
| <rect x="550.2" y="731.0" width="12" height="12" rx="2" fill="#6EA8F2"/> | |
| <line x1="442.3" y1="754" x2="965.0" y2="754" stroke="#0B2A54" stroke-width="2"/> | |
| <path d="M 842.8 747.0 L 849.8 754.0 L 842.8 761.0 L 835.8 754.0 Z" fill="#0B2A54"/> | |
| <text class="value" x="1040" y="742">2.05%</text> | |
| <text class="value" x="1125" y="742">3.56%</text> | |
| <text class="value" x="1210" y="742">8.08%</text> | |
| <rect x="1020" y="796" width="330" height="64" rx="10" fill="#EAF2FF"/> | |
| <text class="small" x="1038" y="821">Supertonic 8-step / Whisper is an insertion-heavy</text> | |
| <text class="small" x="1038" y="841">ASR outlier; Qwen, Nemotron, and listening disagree.</text> | |
| <text class="small" x="56" y="870">95% matched-prompt bootstrap intervals. Formatting-equivalent number forms are normalized before scoring.</text> | |
| </svg> | |