A few years after the voltage regulator that learned to talk, I was having beers with a group of fellow engineers — the kind of informal gathering where the best technical conversations happen.

One of them had recently started his own industrial automation business. I told him about the speech synthesis work I'd done for the trade show. He saw an opportunity immediately.

"Could you build something similar that connects to a PLC?"

The Problem Worth Solving

In industrial environments of that era, process monitoring was largely local. Remote access to alarm conditions — the kind that gets a maintenance technician moving before something fails — was limited or nonexistent.

My friend's idea was to use the existing analog trunked radio infrastructure already present in most factories to deliver voice alarm messages to specific personnel. The system would sit between the PLC and the radio, translating digital alarm codes into spoken alerts directed at the right person.

It was a real problem with a real potential market. That's usually where worthwhile engineering starts.

The Prototype

For the proof of concept, we kept it simple by design. A parallel I/O interface — similar to what I had already built for the voltage regulator development tool — connected the system to the PLC's output ports. The same MC8748 microcontroller I'd used before handled the logic. A handful of alert phrases was all the client needed to validate the concept.

Reusing proven hardware and firmware wasn't laziness — it was the right call for a proof of concept. The goal wasn't to build the final product. It was to answer one question: does this solve a real problem in a real environment?

The prototype went into a mid-size company's control room. My future business partner and his wife — a recently graduated marketing professional — designed a structured evaluation: user feedback, acceptance metrics, a broader market assessment.

What the Market Told Us

The concept worked. Acceptance was positive. But one piece of feedback came up consistently enough that it couldn't be ignored:

The synthesized voice didn't convey urgency. It didn't create an emotional connection.

Allophone-based synthesis — stringing together phonetic building blocks — produces speech that is intelligible but mechanical. For a voltage regulator alerting an operator in a quiet office, that's acceptable. For an alarm system on a factory floor competing with machinery noise and requiring an immediate human response, it falls short.

Market research, as always, had a limitation: people say they'll buy things until they have to write the check. But the proof of concept feedback was different — it came from actual use in an actual environment. That data was reliable.

The implementation had to change. The concept was sound. The technology wasn't.

The Right Reference Point

I've always been drawn to science fiction. The benchmark I had in mind wasn't HAL 9000's unsettling calm — it was the ship's computer in Alien. Authoritative. Precise. Human enough to be understood instantly, but without the condescension of over-designed friendliness.

That's what an industrial alarm voice needs to be. Not a novelty. A tool.

Finding the Technology

My background in audio gave me a starting point. My undergraduate thesis had involved recording digital information on analog video tape — I'd studied PCM encoding, compression methods, bandwidth reduction techniques.

I knew that storing full PCM audio in an embedded system of that era would be prohibitively expensive. But I also knew that bandwidth reduction techniques existed. One in particular — ADPCM (Adaptive Differential Pulse Code Modulation) — had been used in telephony to reduce 64 Kbps PCM streams to 32 Kbps without significant quality loss.

The answer came through a conversation with a colleague at the electronics research center where I worked. Someone mentioned that ADPCM chips were already being used in telephone answering machines, and that a Japanese company called OKI had specialized in exactly this type of circuit.

Access to that kind of information required knowing where to look. We were subscribed to trade journals — EETimes, EDN, Electronic Design — and to the Thomas Register. A fax to OKI later, I had their complete ADPCM databook family on my desk within a month.

The Vision Takes Shape

What OKI offered changed everything. Their chips were capable of storing pre-recorded voice phrases in external EPROM with 4-bit samples at very low data rates — with audio quality more than sufficient for voice.

More importantly, the architecture was right: a microcontroller managing a coprocessor that handled all audio playback autonomously, with the MCU simply queuing phrases and monitoring playback status.

Instead of assembling words from phonemes, we could record complete phrases spoken by a real human voice. That single shift — from synthesis to playback — solved the urgency problem entirely.

The concept for a real product was now clear. Not a one-off prototype. A platform.