Optophone | Hacker News Reader

Optophone

en.wikipedia.org

89 pointsHooke16 comments

The concept of measuring how much ink appears as the text passes a vertical slot came back again in the 1950s. MICR codes, the numbers that appear on checks, are read that way. [1] Or at least were in the original implementation. The ink was magnetized and the paper went past a one-track magnetic tape head. The waveform for each symbol is unique. The recognizer is more like a bar code reader than an OCR system.

There are only 14 characters in that font - the digits 0-9 and four special field identification symbols. The 1970s "futuristic" text fonts which look like MICR symbols are purely decorative.

[1] https://en.wikipedia.org/wiki/Magnetic_ink_character_recogni...

Fascinating, I never realised that's how they work. I found a list of characters and their waveforms. [1]

[1] https://smartcheque.com.au/general-info/about-micr/micr-char...

After reading Hail Mary, I wondered how reasonable it was for someone to truly be able to understand a language based in tones / chords alone. Maybe 60 words per minute would be enough to communicate but it sure would be frustrating.

I think you could get faster with a language actually meant to be 'sung' instead of this rough translation of english characters into audio.

My first thought was: “oh, that’s an interesting concept, I wonder how hard it would be to learn?”

Then I saw the frequency/time graph, and realised that didn’t seem to have been a consideration at all. This was obviously designed by a sighted person who cared more about what the pictures looked like!

Blind person: “But how do I know which letter is which?” Designer: “Oh, that’s easy! Just look at the picture!”

I love the idea of a sung language, though!

Take a look at when this was invented, it's a critical detail in evaluating all this, it was 1913! They were working with the very limited technology they had, they couldn't detect the letters and map them to a particular new tone or chord that might be easier to understand, that tech just wasn't possible [0]. They had to directly translate the image of the letters on simple photo receptors into a corresponding frequency value.

[0] As I was writing this I did have the wild thought that in theory if you had the weights already you could, in theory, implement a very basic character recognition neural net with analog circuitry using vacuum tubes that could recognize letters for direct mapping to sound but it's entirely impractical to create from scratch in reasonable time frames. Maybe over the span of decades you could manually tune one?

IIRC from reading the paper years ago, they chose the tones for each row in the column so they would provide distinct combinations of concordant and discordant sounds.

You might be interested to read about whistled languages, which is pretty close to that idea:

https://en.wikipedia.org/wiki/Whistled_language

Later models of the Optophone allowed speeds of up to 60 words per minute, though only some subjects are able to achieve this rate

Looking at the speeds with which people can communicate with Morse, I suspect that the skill of effectively turning your brain into a UART is something that improves with much practice.

There are some modern systems that convert whole frames from a camera to sound. Apparently some people have learned to interpret information from them but it seems impractically difficult. E.g.

https://www.seeingwithsound.com/

This must be the one I remember playing with around 2010. It used to be a java applet and you could draw in it like a paint program to make interesting sounds. After messing with it for a while I could kind of tell what was going on but not really.

Kind of reminds me of Slow Scan TV - I feel like with enough practice, you could tell what the image looked like without using a decoder.

Is this a lighthearted jab at computer vision being reduced to tokens?

Any simulators so I can listen to how that sounds?

This blog has a recording of the word 'him': https://mallonbacka.com/blog/2023/03/optophone/

I take it this was before speak and spell

This was before integrated circuits and was all analog.

Reply on news.ycombinator.com