Skip to content
← The Workbook

No. 20 · 2026Built

AI Speech Assessor

I have trouble speaking in public. So I built a tool that transcribes audio from my speeches, pulls out filler words, tracks pacing, and suggests improvements. A self-improvement loop I can run before the next speech. Soon to be paired with the wrist-mounted dog-collar from Red Phone v.2. But instead of a shock, a haptic buzz. The polite sibling of electro-therapy: calibrated for low-bar correction (filler words) rather than high-bar (foul language).


Status
Built
Year
2026
Stack
AI · Self-Improvement · Speech · Behavioral · Personal

The Personal Angle

Speaking in public is hard for me. Filler words. "um," "uh," "like," "you know". Creep in as a nervous tic, and the more anxious I get, the faster they multiply. The tool addresses the problem the way I tend to address most problems: build something that measures it, and see if measurement helps.

How It Works

The tool ingests audio of my speech, live or recorded, and produces a structured analysis:

  • Filler word counts by category. Broken down by frequency, position in the speech, and clusters where things spiraled
  • Pacing. Words per minute, pause distribution, sentence length curves
  • Repetition and circular phrasing flagged
  • Lost-thread moments surfaced. Places where the next sentence didn't actually follow the previous one
  • Suggested rewrites for the parts that didn't land

The report is what I read before the next speech. Studying my own bad habits is uncomfortable. It's also the fastest way to fix them.

What's Next

The next iteration pairs the assesser with the wrist-mounted dog collar I'm building for the Red Phone v.2 (Electro Therapy). Instead of a shock, the vibrate mode. A small, polite tap on the wrist when filler patterns start firing. The shock collar's gentlest setting, repurposed.

The loop tightens: speak → real-time analysis → buzz → adjust. Operant conditioning for fluency, dialed softer than the foul-language version. The body learns what the report told the brain.

Why It Pairs With the Red Phone v.2

Same hardware. Same philosophy. Different threshold. The v.2 was built for high-bar correction (rudeness, foul language). The Speech Assessor is for low-bar correction (filler words). Both use my own behavior data against me, applied through the body in real time. The collar is a single-purpose tool with a tunable threshold; the right threshold for "uh" is gentler than the right threshold for "fuck."

Status

Built. Offline analysis runs. Haptic-buzz integration planned.