assets/evidence/quality-vs-footprint.svg

<svg xmlns="http://www.w3.org/2000/svg" width="1500" height="730" viewBox="0 0 1500 730">
<rect width="1500" height="730" 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="730" rx="4" fill="#1769E0"/>
<text class="title" x="56" y="52">Predicted quality versus model footprint</text>
<text class="subtitle" x="56" y="78">UTMOS22 on 500 matched unseen prompts · complete deployable weight footprint</text>
<line class="grid" x1="157.8" y1="125" x2="157.8" y2="610"/>
<text class="small" x="157.8" y="638" text-anchor="middle">16</text>
<line class="grid" x1="297.2" y1="125" x2="297.2" y2="610"/>
<text class="small" x="297.2" y="638" text-anchor="middle">32</text>
<line class="grid" x1="436.6" y1="125" x2="436.6" y2="610"/>
<text class="small" x="436.6" y="638" text-anchor="middle">64</text>
<line class="grid" x1="576.0" y1="125" x2="576.0" y2="610"/>
<text class="small" x="576.0" y="638" text-anchor="middle">128</text>
<line class="grid" x1="715.4" y1="125" x2="715.4" y2="610"/>
<text class="small" x="715.4" y="638" text-anchor="middle">256</text>
<line class="grid" x1="854.8" y1="125" x2="854.8" y2="610"/>
<text class="small" x="854.8" y="638" text-anchor="middle">512</text>
<line class="grid" x1="100" y1="610.0" x2="850" y2="610.0"/>
<text class="small" x="86" y="614.0" text-anchor="end">4.12</text>
<line class="grid" x1="100" y1="488.8" x2="850" y2="488.8"/>
<text class="small" x="86" y="492.8" text-anchor="end">4.21</text>
<line class="grid" x1="100" y1="367.5" x2="850" y2="367.5"/>
<text class="small" x="86" y="371.5" text-anchor="end">4.29</text>
<line class="grid" x1="100" y1="246.2" x2="850" y2="246.2"/>
<text class="small" x="86" y="250.2" text-anchor="end">4.38</text>
<line class="grid" x1="100" y1="125.0" x2="850" y2="125.0"/>
<text class="small" x="86" y="129.0" text-anchor="end">4.46</text>
<circle cx="329.3" cy="217.7" r="13" fill="#1769E0"/>
<text x="329.3" y="221.7" text-anchor="middle" font-size="11" font-weight="800" fill="#FFFFFF">1</text>
<circle cx="157.5" cy="230.4" r="13" fill="#1769E0"/>
<text x="157.5" y="234.4" text-anchor="middle" font-size="11" font-weight="800" fill="#FFFFFF">2</text>
<circle cx="804.2" cy="360.8" r="13" fill="#8B9AA9"/>
<text x="804.2" y="364.8" text-anchor="middle" font-size="11" font-weight="800" fill="#FFFFFF">3</text>
<line x1="433.8" y1="503.0" x2="433.8" y2="368.4" stroke="#8B9AA9" stroke-width="3"/>
<circle cx="433.8" cy="435.7" r="13" fill="#8B9AA9"/>
<text x="433.8" y="439.7" text-anchor="middle" font-size="11" font-weight="800" fill="#FFFFFF">4</text>
<line x1="412.5" y1="508.6" x2="412.5" y2="471.9" stroke="#8B9AA9" stroke-width="3"/>
<circle cx="412.5" cy="490.2" r="13" fill="#8B9AA9"/>
<text x="412.5" y="494.2" text-anchor="middle" font-size="11" font-weight="800" fill="#FFFFFF">5</text>
<rect x="905" y="116" width="540" height="420" rx="12" fill="#FFFFFF" stroke="#DCE6F1"/>
<text class="small" x="927" y="146">MODEL FAMILY</text>
<text class="small" x="1325" y="146" text-anchor="end">UTMOS22</text>
<text class="small" x="1415" y="146" text-anchor="end">MB</text>
<circle cx="927" cy="174" r="11" fill="#1769E0"/>
<text x="927" y="178" text-anchor="middle" font-size="10" font-weight="800" fill="#FFFFFF">1</text>
<text class="label" x="949" y="178">Inflect-Micro-v2</text>
<text class="value" x="1325" y="178" text-anchor="end">4.395</text>
<text class="small" x="1415" y="178" text-anchor="end">37.5</text>
<circle cx="927" cy="240" r="11" fill="#1769E0"/>
<text x="927" y="244" text-anchor="middle" font-size="10" font-weight="800" fill="#FFFFFF">2</text>
<text class="label" x="949" y="244">Inflect-Nano-v2</text>
<text class="value" x="1325" y="244" text-anchor="end">4.386</text>
<text class="small" x="1415" y="244" text-anchor="end">16.0</text>
<circle cx="927" cy="306" r="11" fill="#8B9AA9"/>
<text x="927" y="310" text-anchor="middle" font-size="10" font-weight="800" fill="#FFFFFF">3</text>
<text class="label" x="949" y="310">Supertonic 3 · 8-step</text>
<text class="value" x="1325" y="310" text-anchor="end">4.295</text>
<text class="small" x="1415" y="310" text-anchor="end">398.1</text>
<circle cx="927" cy="372" r="11" fill="#8B9AA9"/>
<text x="927" y="376" text-anchor="middle" font-size="10" font-weight="800" fill="#FFFFFF">4</text>
<text class="label" x="949" y="376">Piper Low · 2-voice mean</text>
<text class="value" x="1325" y="376" text-anchor="end">4.242</text>
<text class="small" x="1415" y="376" text-anchor="end">63.1</text>
<circle cx="927" cy="438" r="11" fill="#8B9AA9"/>
<text x="927" y="442" text-anchor="middle" font-size="10" font-weight="800" fill="#FFFFFF">5</text>
<text class="label" x="949" y="442">KittenTTS Nano · 2-voice mean</text>
<text class="value" x="1325" y="442" text-anchor="end">4.204</text>
<text class="small" x="1415" y="442" text-anchor="end">56.8</text>
<text class="small" x="475.0" y="675" text-anchor="middle">Model weight footprint (MB, logarithmic scale)</text>
<text class="small" x="26" y="370" transform="rotate(-90 26 370)" text-anchor="middle">Predicted MOS · higher is better</text>
<text class="small" x="56" y="710">UTMOS22 is an automated predictor, not human MOS. Family means use equal weighting across the named voices.</text>
<text class="small" x="905" y="565">Supertonic 3-step: 2.471 UTMOS22 · below plotted range.</text>
<text class="small" x="905" y="585">Supertonic UTMOS uses the James voice; family means use equal voice weighting.</text>
</svg>