Get amazing AI audio voiceovers made for long-form content such as podcasts, presentations and social media. (Get started for free)

Robocaller 2.0: I Built a GPT to Make Phone Calls (And Prank My Friends)

Robocaller 2.0: I Built a GPT to Make Phone Calls (And Prank My Friends) - Finding the Perfect Voice Samples

One of the most important parts of training an AI to clone a voice is gathering high-quality voice samples. The better the source material, the more accurate the final cloned voice will be. When I set out to build my AI phone prankster, I knew I had to be choosy about whose voice I cloned.

I considered using celebrity voices, but quickly realized I would never be able to gather enough public samples to properly train the AI. I needed at least a few minutes of clean audio, preferably from one continuous recording so the AI could learn the natural cadences. Celebrities rarely release long, unedited voice samples.

Another option was cloning my own voice or a friend's, but that felt too predictable. I wanted a voice that would really confuse my prank targets. So I turned to casting calls and voice acting websites, places where aspiring talents post samples of their work.

Sifting through the demos, I listened for voices that were clearly audible and free of background noise. I avoided samples with music or sound effects. The rawer the better. I also looked for a unique vocal quality that would be tricky to place. Too squeaky or too deep and my friends would guess it was fake. I needed that sweet spot right in the middle.

After a couple hours of searching, I found it - a smooth mid-Atlantic accent from a voice actor named Sam. He had posted a few commercial-style reads that showed off his versatility. I ran the samples through audio editing software to remove any clicks, pops, or artifacts.

Chopped up and combined, I was able to build a three minute sample that contained enough phonetic diversity to properly train the AI. It had all the basic sounds - the vowels, consonants, noises, laughs. I even snipped little silences and breaths to capture the natural rhythm of Sam's speech.

With a cleanly edited voice sample, I could finally start the training process. Having high quality source material made a world of difference compared to my early failed experiments with noisy audio. The clarity enabled the AI to learn Sam's cadence and tone with surprising accuracy.

Of course, gathering voice samples ethically and legally is an important consideration when cloning voices with AI. I made sure Sam's demos were publicly available with commercial use allowed. I would advise others exploring this technology to be similarly careful about permission and usage rights.

Robocaller 2.0: I Built a GPT to Make Phone Calls (And Prank My Friends) - Building a User-Friendly Interface

A key part of bringing any new technology to life is designing an intuitive user interface. For my DIY robocaller bot, I knew that building a streamlined UI would be crucial. The interface needed to be simple enough that anyone could use it, not just engineers. My goal was enabling friends and family to prank each other without needing to understand the complex AI under the hood.

The interface design process started on paper. I sketched out different options for allowing users to input a phone number, record a custom voicemail message, and trigger the call. Early versions involved too many steps and fields. I wanted to boil it down to the bare essentials.

After refining the workflow, I moved to coding the UI itself. I built it as a web app so it could work on phones, tablets, laptops - anything with a browser. The framework handled cross-device responsiveness so I could focus on usability.

On the design side, I kept the interface clean and uncluttered. Each element had plenty of breathing room. I chose an inviting background color and made buttons large and clickable. The goal was to eliminate any point of friction that might confuse users.

For entering phone numbers, I provided country code dropdowns and formatted inputs. That prevented mistakes from incorrect or missing dial codes. I also used validation to catch any non-numeric entries.

Recording custom voicemail messages posed an interface challenge. I needed an intuitive recorder that averaged users could figure out. My solution was a large round button to start and stop recording. I displayed sound waveforms as visual feedback. For playback, I used a similar round button marked with a triangle.

With the interface built, I started testing. I asked friends and family to try making calls. When they struggled at any point, I made tweaks and simplifications. Their feedback really refined the workflow.

By the end, I had an interface that felt like second nature. Users could log in, enter a number, record a message, and fire off the robocall prank with just a few taps. The complexity of the underlying AI was entirely abstracted away.

That simplicity enabled anyone to unleash the bot without technical expertise. The result was hilarious prank calls spreading like wildfire among friends. But it also made me reflect on how easy it is now for people to misuse AI. With the right interface, powerful technologies become incredibly accessible.

Robocaller 2.0: I Built a GPT to Make Phone Calls (And Prank My Friends) - Unleashing the Bot on Unsuspecting Friends

When I felt my robocaller bot was finally ready, it came time for the fun part - unleashing it to prank my friends. I started by picking a close friend I knew would find the joke hilarious, not creepy. Still, I seeded the rollout by giving them a vague heads up: "Hey, you may get a weird call today, just wanted to warn you!" That primed their expectations for something odd without ruining the surprise.

With my first target selected, I logged into the web interface on my phone, entered their number, and recorded a short message saying, "Hey [Friend's Name], this is [Other Friend] calling to see if you wanted to get dinner tonight? Let me know!" I hit send and waited eagerly.

Sure enough, a few minutes later, I got a confused call from my friend asking if I had just left the voicemail. They were perplexed, knowing I hadn't actually called but hearing a voice that sounded convincingly like mine. I played dumb before finally revealing it was a robocall prank, at which point we both had a good laugh. My friend demanded I send them the web interface link so they could prank others.

This first success gave me confidence to start gradually expanding the bot's reach. I moved on to less familiar acquaintances that I thought would take the joke well. Their reactions ranged from utter bewilderment to laughing hysterics. Many begged for the web link to prank their own friends and family.

When my sister called my mom using the robocaller, my mom was completely fooled. She went off on my sister for a good 5 minutes about ignoring her calls. When my sister revealed it was a prank, my mom was amazed by the technology but also saw the danger of it being misused.

Others had more wary reactions, especially to calls from unknown numbers. A few friends felt deeply unsettled by how real the voices sounded until I explained it was AI. This made me step back and consider the ethics of using such deceivingly human-like voices for pranks without proper consent.

While most friends embraced the prank after understanding it was harmless fun, I realized I had to be more selective with targets. Spoofing close friends was one thing, but tricking strangers or spamming numbers crossed a line. I began limiting the bot's scope to people who expressly permitted it.

Robocaller 2.0: I Built a GPT to Make Phone Calls (And Prank My Friends) - The Prank War Begins - Friends Fight Back

Once my inner circle got their hands on the robocall bot interface, all hell broke loose. What started as some harmless fun rapidly escalated into an all-out prank war. Friends were ruthlessly spoofing each other's voices to tell ridiculous lies or share embarrassing secrets. No one felt safe from the next prank call.

My buddy James woke up to a voicemail from his mom scolding him for dropping out of college. But the twist was that it wasn't actually his mom - it was our friend Lily using the robocaller to perfectly mimic his mom's voice. James was furious until he realized it was a prank. Then he used the bot to get revenge by leaving Lily a fake voicemail from her crush asking her on a date.

Another friend group staged an intervention for their friend Katie using the robocaller. Each person recorded a message expressing concern over Katie's "drinking problem" and urging her to get help. Katie was moved to tears by her caring friends until someone finally revealed it was all a trick. She promptly used the bot to call each friend and cuss them out in their own voices!

Some of the more tech-savvy friends even got creative with audio editing software to splice together different words and phrases from the bot to say new things. Sentences could be constructed by mixing and matching sounds. This let them put words in peoples' mouths that were never actually said.

By far the most elaborate prank came from my friend Rachel who used the robocaller to slowly gaslight her roommate. It started with a few odd calls mimicking acquaintances saying they'd be late to meet up. Then came a fake call from the landlord about unpaid rent. After weeks of bizarre calls, the roommate was questioning her sanity, until Rachel finally revealed the escalating prank.

While I intended the robocaller to be harmless fun, things were starting to get out of hand. Friends justified increasingly bold lies by saying it wasn't really them making the call. The anonymity enabled bad behavior. I realized then that I had greatly underestimated the potential for abuse and deception.