speech vs brain

speech-vs-brain.png

Output file: images/speech-vs-brain.png
Title: Thin client split

Copy everything inside the fenced block below into Gemini (Imagen). Save the result as speech-vs-brain.png in the wiki images/ folder.

Visual style for COMSTAR AI wiki documentation:
- Dark charcoal / near-black background (#0B0E12 to #141A22)
- Accent colors: cool cyan (#3DDCFF) and warm amber (#F0A202) only — no purple, no neon magenta
- Clean technical diagram aesthetic: crisp vector-like shapes, subtle grid or soft radial vignette
- Typography in-image: geometric sans, sparse labels, high contrast white/light gray text
- Mood: engineered home AI terminal — precise, calm, slightly industrial, not sci-fi hologram cliché
- No photorealistic faces of private individuals; use silhouettes or generic figures when people appear
- No logos of third-party companies unless abstractly represented as labeled boxes
- Aspect ratio: 16:9 unless noted
- High resolution, sharp edges, suitable for GitHub wiki (readable at ~900px wide)
- Leave modest padding around edges; avoid tiny unreadable text

Split diagram: left "I/O on Pi" (mic waveform, wake word, VAD, kiosk playback); right "Compute on Ada" (STT, TTS, AO planner, YOLO/face). Center arrow "LAN only". Emphasize PCM and JPEG crossing to the right; transcripts returning.

Avoid: purple gradients, stock "AI brain" tropes, excessive glow bloom, cluttered dashboards, comic-book style, watermarks, illegible micro-text, photoreal private faces, corporate Memphis patterns, cream terracotta aesthetic