My AI Voice

 


I have an AI voice.

It's not my regular voice -  It's the voice I put on when speech-to-text is enabled and I'm about to give instructions to Claude, ChatGPT, or whatever AI I've summoned that day. It emerges automatically, like a reflex. Like how people talk louder on the phone for no reason.

The AI voice has many variations. There's the Casual One, the Authoritative One, the Slightly Too Loud One (for when I'm in public and pretending I'm not talking to a computer). There's even a Polite One, because apparently I was raised to say please to a language model - eats up compute time I've heard. 

None of them work very well.

The ones that do work — the ones where the AI actually understands me — all share one thing in common: they sound robotic. Slow. Deliberate. Slightly dead inside. Like I'm reading terms and conditions out loud to a jury.

"Please. Write. A. Blog. Post. About. My. AI. Voice."

It works great. The transcription is perfect. I sound like I've never met a human before.

I'm trying to fix this. I want to sound natural and be understood. I want to flow. I want warmth. I want to not feel like I'm leaving a voicemail for someone I've disappointed.

So far, every attempt to sound more natural has produced creative new ways to confuse the AI. I said "you know what I mean?" once and it transcribed it as a question and waited for an answer.

I'll keep working on it. In the meantime, if you hear someone in a coffee shop speaking in a calm, measured monotone to no one in particular —

That's me. Nailing it.

PS: Update, have experiemented with accents, Indian, Scottish, Irish, South SIde gangster, Chinese, Louisiana drawl, however faux BBC with discernable pauses is the current champion. 

Comments

Post a Comment

Popular posts from this blog

Building AI-Readable UI: The Next Frontier for Frontend Engineers

Image
Building AI-Readable UI: The Next Frontier for Frontend Engineers The way we build user interfaces is a moving target — not because humans are interacting with them differently, but because AI agents are now sitting between humans and the software they use. The release of Claude in Chrome is a case in point - operating directly in the browser, parsing the DOM - Claude has got its little claws in there. (you see what I did there?) But it sees you the way a screen reader does, not the way a human does. It reads your HTML. It traverses your structure. It infers meaning from from the DOM, from the labels, roles, and attributes. And here's the uncomfortable bit – we’ve been feeding it rubbish. The Looming Problem For the past decade, the frontend world has optimised for visual output. We moved to component-based frameworks, CSS-in-JS, and design systems that produce beautiful, pixel-perfect interfaces. But we may have lost the semantic scaffolding that makes the DOM language understanda...
Read more

The Forgotten UI

Image
Admins are striking for better UI - True Story There's a pattern in frontend development worth talking about. We invest deeply in the consumer-facing experience, the hero section gets the love, the typography, the sexy micro-animations. But the admin UI wher content editors live for six hours a day, get the bums rush..A data table, a save button, and we're moving on. The homepage is visible, demonstrable, impressive in a client presentation. The backend is "she'll be right mate" until it's not. The Admin UI Is a Workplace Content editors do knowledge work - inventory, copy, publishing, customer records , under pressure, often without technical support nearby. The quality of the UI they work in directly affects the quality and speed of their output. State management matters here more than almost anywhere else. Users need to know what's saved, what's draft, what's published. Did the image upload? Was the email sent? Feedback on every action isn'...
Read more

The Great Software Collapse

Image
The Great Software Liberation: Why Every SaaS Will Be Free There's a shift coming most enterprise software companies wont appreciate. Not a gradual evolution, more a collapse.Like 3,000 video editing suites when iMovie shipped for free, or GPS hardware when Google Maps arrived on your phone.  Hollywood movie title altert : "T his time is it's coming for everything, all at once" Vibe Coding - The New App Store Cast your weary mind back to 2008 - oh wait you may not have been born yet. The App Store launched and within a few years, entire software categories were wiped out or permanently repriced. A $400 standalone GPS became a free app overnight.  AI-assisted vibe coding is that moment, but for enterprise software. Tools like Lovable, Cursor, and Bolt let a non-engineer describe what they want and have a working application running in minutes on Postgres via Neon or Supabase with instant production access. The barrier between "I have an idea" and "I have...
Read more