The Man Who Compressed the Human Voice

It's Friday, September fourth. Quick thing before we get into today's story: if there's ever a topic you want me to dig into, there's a feedback option right there in the HigherIQ app. Leave it there and I'll see it. That is actually how today's episode came about. My dad texted me last night about something he had just read, and it was interesting enough that I dropped everything else and went down the rabbit hole myself.

Every mobile phone call you have ever made was compressed before it left your phone. Your voice does not travel. A mathematical description of your voice travels, and the phone at the other end rebuilds it. The man who figured out how to do that was born in Kanpur, India. His name is Bishnu (BISH-noo) Atal. He died two weeks ago on August nineteenth at the age of ninety-three, and almost nobody noticed.

The problem he worked on sounds dull until you understand what it prevents. Sending a voice as raw sound uses an enormous amount of bandwidth. When mobile phones started spreading in the nineteen eighties, the networks began to fill up. There is only so much radio spectrum. You cannot manufacture more of it. Without compression, mobile telephony would have stayed a luxury for a tiny number of people. There simply would not have been enough room in the air for everyone.

Think about what that means. The thing that made mobile phones accessible to billions of people was not cheaper hardware or better batteries. Those helped. But the fundamental constraint was physics. Every call occupies a slice of the electromagnetic spectrum, and that slice is finite and shared. The only way to fit more calls into the same slice is to make each call smaller. Someone had to figure out how to shrink a human voice without destroying it.

At the time Atal started working on this, in the mid nineteen sixties, the idea of carrying a phone in your pocket was science fiction. Bell Labs, where he worked, was the most productive research laboratory in the world. It had already produced the transistor, the laser, the solar cell, and the Unix operating system. Atal walked into that building in nineteen sixty-one and stayed for thirty-five years. His assignment was simple to describe and brutally hard to solve: take a human voice and compress it so tightly that you can send it through a straw.

Between nineteen sixty-six and nineteen seventy, Atal developed something called linear predictive coding, or LPC. The idea is elegant. Human speech is highly predictable. The sound you are making right now is closely related to the sound you made a fraction of a second ago, because it is produced by the same throat and the same mouth shape. Your vocal tract is a physical system that changes slowly. It does not jump randomly from one sound to the next.

So instead of sending the sound itself, you send a small set of numbers that describe how the sound is being produced. Parameters for the shape of the vocal tract, for whether the sound is voiced or unvoiced, for the pitch. The receiver uses those numbers to reconstruct the speech from scratch, like a synthesizer being told what to play.

Atal's early version worked at just twenty-four hundred bits per second. To give you a sense of how small that is, uncompressed CD-quality audio is about one point four million bits per second. Atal's system was compressing the signal by a factor of nearly six hundred. The American military adopted it for secure telephones during the Cold War. It worked. You could understand the words. But it did not sound like a person. It was intelligible and slightly robotic. You would know you were talking to a machine.

That was useful for the military. It was not good enough for everyone else. Nobody wants to call their mother and sound like a robot. Atal knew that, and he spent the next fifteen years trying to fix it.

In nineteen eighty-five, working with a German physicist named Manfred Schroeder, Atal published a method called code-excited linear prediction, or CELP. The CELP approach added a crucial piece. It stored a codebook of typical excitation signals, thousands of tiny waveforms that represent different ways the human voice can sound. For each short segment of speech, the encoder searches through that codebook, finds the waveform that produces the closest match to the original, and sends only the index number. The decoder looks up the same waveform from the same codebook and uses it to reconstruct natural-sounding speech.

What Atal had essentially done was build a catalog of the building blocks of human speech, and then figure out how to describe any voice using only references to that catalog. The result was compression ratios that seemed impossible, combined with voice quality that sounded like a real person. You could not tell the compressed voice from the original.

This is the compression method inside practically every digital cellular standard in the world. GSM. CDMA. Voice calls over the internet. When you make a phone call today, CELP or one of its direct descendants is almost certainly compressing your voice before it leaves your device. The method also enabled wireless networks in developing countries that lacked cable infrastructure. You did not need to lay copper across a continent when you could fit so many calls into so little spectrum.

Atal holds sixteen American patents and many international ones. The National Academy of Engineering elected him in nineteen eighty-seven. Bell Labs made him a fellow in nineteen ninety-four. In two thousand three, the Franklin Institute gave him the Benjamin Franklin Medal in electrical engineering. The citation they wrote was one sentence. "The next time you make a call on your cell phone, think of Bishnu Atal."

Almost nobody does. I had never heard his name before my dad sent me that text. He was born in Kanpur, took a physics degree at the University of Lucknow at nineteen, got a diploma at the Indian Institute of Science in Bangalore, taught acoustics there for three years, then flew to New Jersey and spent the next thirty-five years solving a problem that determined whether mobile phones would be for the few or for everyone. He died on August nineteenth, two weeks ago, at ninety-three. Indian news agencies did not cover it. The gap between the scale of what he made possible and the scale of the recognition he received is something I have not been able to stop thinking about. So thanks, Dad, for sending me that text. Stay informed, stay curious, and we'll see you on Monday.