Daniel sent us this one — he wants us to dig into the people in courtrooms who are tapping away on those strange little machines, stenographers. Not to be confused with steganographers, which as he points out, are the folks hiding information in plain sight. He studied law back in Cork, watched district court hearings, and assumed like most people that court reporters were just outrageously fast typists on regular keyboards. Turns out, completely different device, completely different method. They type phonetically, not letter by letter. And Daniel's asking: in a world where Whisper and automated speech recognition are getting extremely accurate, is this profession moving toward editing imperfect transcripts, or is it still faster to just do it manually from the start? Where else do these people work? How long does it take to learn? And can you just buy one of these machines and become the guy at the coffee shop with the weirdest keyboard anyone's ever seen?
I love this topic. And by the way, DeepSeek V four Pro is writing our script today, so we'll see if it can keep up with stenographic speeds.
I assume it types faster than I talk, which is admittedly a low bar.
Let's start with what most people get wrong, because you hinted at it. A stenotype machine is not a keyboard. It has twenty-two keys, sometimes a few more on modern digital versions, and you press multiple keys at once — chords, like a piano. You're not typing letters sequentially. You're pressing combinations that represent syllables, whole words, sometimes entire phrases in a single stroke.
The hundred-words-per-minute touch typist Daniel mentioned? That's not even in the same league.
A good QWERTY typist maxes out around a hundred twenty words per minute if they're exceptional. The certification standard for court reporters in the United States, set by the National Court Reporters Association, requires two hundred twenty-five words per minute for literary dictation, two hundred words per minute for jury charge, and a hundred eighty words per minute for testimony — which involves multiple speakers, interruptions, overlapping dialogue. And those are the minimums. Many experienced stenographers hit two sixty, two eighty, even three hundred words per minute in bursts.
That is genuinely hard to wrap your head around. Three hundred words per minute. The average person speaks at about one hundred fifty. So a stenographer at full speed is capturing speech faster than it's being produced, which means they can handle the most rapid-fire cross-examination without dropping a word.
Here's the thing about how they do it. The stenotype machine uses a phonetic system. The keys are arranged in a specific layout — there's an initial consonant bank on the left, vowels in the middle using the thumbs, and final consonants on the right. A single stroke might combine S, T, K, and a vowel to produce the sound "stak." The machine doesn't output English text directly. The raw output is called steno code, and it looks like absolute gibberish to the untrained eye.
Give me an example. What does "steno code" actually look like?
The word "stenographer" might come out as something like "STEPB GRAF ER." The software on the back end — called computer-aided transcription — translates those phonetic strokes into English using a personal dictionary that the stenographer has built over years. Every stenographer's dictionary is customized. If you add a new word, a technical term, a proper name, you define its steno stroke once, and the computer recognizes it thereafter.
There's a whole translation layer happening in real time. That explains why it's not just about fast fingers. You're mentally decomposing speech into phonemes, mapping those to chord combinations, and trusting your custom dictionary to reassemble everything correctly on the fly.
And the training timeline reflects that. Becoming a certified court reporter typically takes two to four years of intensive study. The dropout rate is enormous — some estimates put it at eighty-five to ninety percent. It's not the theory that gets people. The theory of steno, you can learn in a few months. It's building the muscle memory to execute two hundred twenty-five words per minute with ninety-five percent accuracy or better. That's years of daily practice.
Daniel asked where else these people work beyond courtrooms. I assume it's not just legal proceedings.
Far from it. Broadcast captioning is enormous. Every live television news program, every sporting event, every awards show — someone is writing real-time captions, often from a remote location. The stenographer listens to the broadcast feed and their steno output gets encoded into the closed captioning data stream. The delay is typically two to three seconds. If you've ever watched live captions during a breaking news event and marveled at how fast they appear, that's a stenographer.
I've also seen them at university lectures for deaf or hard-of-hearing students. There's a whole field called Communication Access Realtime Translation, CART. A stenographer sits in the lecture hall, transcribes everything the professor says, and the text appears on the student's laptop in real time.
CART providers also work in business meetings, conferences, religious services, even weddings and funerals. Anywhere someone needs verbatim real-time access to spoken content. The Americans with Disabilities Act created substantial demand for CART services in educational and workplace settings.
There's also closed captioning for theater performances, legislative proceedings — every state legislature and Congress has stenographers or their digital equivalents. Grand jury proceedings, depositions, arbitrations, administrative hearings. The Federal Communications Commission requires closed captioning on essentially all television programming, and while a lot of pre-recorded content uses offline captioning, anything live still requires a human in the loop.
Let's talk about the government context, because this connects back to some of Daniel's earlier prompts about diplomatic note-taking. In Congress, the official record is the Congressional Record, and the people who produce it are official reporters of debate. The House and Senate each have their own teams. They work in fifteen-minute shifts because the cognitive load is so intense. After their shift, they go back and edit their transcript, checking quotes, resolving ambiguities, inserting correct spellings of names and places.
Fifteen-minute shifts. That tells you everything about the mental demands of this job.
It really does. And here's a detail I find fascinating: the stenotype machine itself has evolved enormously. The classic model was the Stenograph company's paper-based machines, which punched holes in a narrow paper tape, same principle as a player piano. The steno code was literally encoded as physical holes. If the computer-aided transcription software failed, you still had the paper tape as a backup.
That feels wonderfully analog. A courtroom's entire official record, reducible to a spool of punched paper.
Modern machines are fully digital. Companies like Stenograph, ProCAT, and Neutrino Group make machines with LCD screens, USB connectivity, Bluetooth. They run translation software in real time. Some are the size of a small laptop. But the core interface — the twenty-two-key chorded layout — hasn't fundamentally changed since the early twentieth century.
Daniel asked whether you can just buy one. I looked into this. You absolutely can. A new professional stenotype machine runs anywhere from two thousand to five thousand dollars. Used machines are available for less. And there is, in fact, a hobbyist community.
The Open Steno Project. This is one of my favorite corners of the internet. It started around twenty-ten when a programmer named Mirabai Knight, who is a professional stenographer, wanted to make steno accessible to anyone. She developed Plover, which is free open-source stenography software that runs on a regular computer. And here's the key innovation: Plover lets you use a regular NKRO keyboard — that's n-key rollover — as a steno input device. You don't need a dedicated stenotype machine.
You can literally remap a mechanical keyboard to function as a steno writer?
Any keyboard that supports n-key rollover — meaning it registers every key press individually no matter how many you press simultaneously — can run steno chords. You press multiple keys at once, Plover interprets the chord, and outputs the corresponding text. There are also purpose-built hobbyist steno keyboards, like the Georgi, the TinyMod, the SOFT/HRUF Splitography. These are open-source hardware designs that typically cost between a hundred and three hundred dollars, a fraction of the professional machines.
This is exactly the niche Daniel was asking about. The person in the coffee shop with the weirdest keyboard.
It absolutely exists. There's a whole community on Discord and Reddit, people who learn steno not because they want to become court reporters, but because they want to type at two hundred words per minute for coding, for writing, for everyday computer use. Some of them use it as an accessibility tool — people with repetitive strain injuries find that chording puts far less strain on their hands than traditional typing.
That makes intuitive sense. Fewer individual key presses, less finger travel, less repetitive motion.
There's a small but passionate subculture of programmers who write code using steno. They create custom dictionaries with steno strokes for common programming syntax — entire function definitions, loops, boilerplate code in a single chord. I saw one developer who could output a full Python class definition in three strokes.
Now I'm imagining a courtroom where the stenographer finishes transcribing a cross-examination and then casually writes a Django application during recess.
The skills are transferable. But let me address Daniel's core question about AI and the future of this profession, because this is where things get interesting.
He mentioned Whisper and automated speech recognition. The question is whether we're moving toward a world where stenographers become editors of AI-generated transcripts rather than primary creators.
The honest answer is that the field is already changing, but not in the straightforward replacement way people assume. Automated speech recognition has gotten remarkably good. Whisper, OpenAI's model, and its successors can transcribe clean audio with word error rates in the single digits. In ideal conditions — clear speech, no overlapping dialogue, limited jargon — it can hit ninety-eight percent accuracy or better.
A courtroom is not ideal conditions.
That's the crux of it. A courtroom has multiple speakers who interrupt each other. It has people with accents, people who mumble, people who speak in incomplete sentences. It has specialized legal terminology, case citations, Latin phrases, proper names of witnesses and defendants and companies that the ASR system has never encountered. It has ambient noise, HVAC systems, papers rustling, doors opening. And most importantly, the transcript is a legal document. Ninety-eight percent accuracy in a business meeting is fine. Ninety-eight percent accuracy in a murder trial transcript is a disaster.
One misheard word changes the entire meaning of testimony. "I did not see the weapon" versus "I did see the weapon.
The legal system requires verbatim accuracy. Not close, not mostly right, not good enough. So what's happening in practice is something of a hybrid model. Some court systems, particularly in states with severe stenographer shortages — and there is a massive shortage, which I'll get to — are experimenting with digital court reporting. This involves high-quality multi-channel audio recording combined with automated transcription, and then a human editor reviews and corrects the transcript.
How does that compare in practice?
The digital court reporters — and I should distinguish them from stenographers, they're a different profession with different certification standards — use specialized recording equipment with multiple microphones, each channel isolated. They take notes during the proceeding, marking timestamps, identifying speakers, flagging difficult passages. After the proceeding, the ASR generates a draft transcript, and the digital reporter reviews it against the audio, correcting errors, formatting, and producing the final certified transcript.
The workflow shifts from real-time capture to post-hoc correction. Is it faster?
That's precisely the debate Daniel alluded to. And his instinct is actually backed up by many professionals in the field. Editing a flawed transcript can be surprisingly slow. When you're correcting someone else's errors — or an AI's errors — you have to listen to the audio, read the text, spot the discrepancy, and fix it. That's a different cognitive process than generating the text yourself. Many stenographers argue that it's faster to just write it correctly the first time than to clean up after an ASR system.
I've experienced this myself, actually. Correcting a long AI transcript, you start thinking, I could have just typed this out from scratch in less time.
There's research on this. The correction overhead can be substantial. If ASR is ninety-five percent accurate, you're fixing one word in twenty. That sounds manageable, but those errors are not randomly distributed. They cluster around proper nouns, technical terms, accented speech — exactly the places where accuracy matters most. And each correction requires a context switch: pause, rewind, listen, compare, edit, resume. It breaks flow.
The human in the loop ends up doing work that may not be faster than the human operating a stenotype machine from the start.
For live proceedings where an immediate transcript is required, stenography remains the gold standard. The real-time aspect matters enormously. A judge can ask the stenographer to read back a portion of testimony immediately. Attorneys rely on real-time feeds during trial to review what a witness just said. In depositions, the real-time transcript lets lawyers adjust their strategy on the fly.
That's a capability ASR alone doesn't provide in the same way. Even with near-instant transcription, the stenographer's output is already verified by a human brain in real time. The ASR output is raw and unchecked.
Now let me talk about the shortage, because it's driving a lot of these conversations. The National Court Reporters Association has been warning about a critical shortage for years. The average age of a court reporter in the US is over fifty. More stenographers are retiring than entering the profession. The number of court reporting programs has shrunk dramatically — from over six hundred programs in the nineteen eighties to fewer than a hundred today. Some states have fewer than ten new certified reporters entering the field each year.
What's causing the decline?
The training is brutally difficult, as I mentioned. The dropout rate is astronomical. Young people don't necessarily see it as an attractive career path — they assume, as Daniel did, that technology will make the profession obsolete. And the pay, while decent, may not justify four years of intensive training for many people. The median salary for court reporters is around sixty to sixty-five thousand dollars, though experienced freelancers in major markets can earn well over a hundred thousand.
You have a profession that's essential to the functioning of the legal system, with a workforce that's aging out and not being replaced fast enough. That's a genuine crisis.
And it's why many jurisdictions are actively exploring alternatives. Digital court reporting is growing rapidly. Some states have created new certification pathways for digital reporters. The American Association of Electronic Reporters and Transcribers, AAERT, certifies digital court reporters with their own standards and exams.
There's also a completely different method we haven't mentioned: voice writing.
Voice writing, or stenomask reporting. Instead of a stenotype machine, the reporter wears a mask that covers their mouth, connected to a highly directional microphone. They repeat everything they hear — every word from every speaker — into the mask. The mask prevents anyone else from hearing them. The audio is fed to speech recognition software that's been trained on the reporter's voice, which is much more accurate than general speaker-independent ASR.
They're essentially acting as a human relay, repeating the testimony verbatim under their breath.
And because the software is trained on one specific voice, accuracy can be extremely high. Voice writers can achieve certification speeds comparable to stenographers. The training period tends to be shorter — typically eighteen months to two years. It's more physically sustainable for some people. The mask looks a bit odd, but it works.
I've seen those in courtroom sketches. I always assumed they were wearing some kind of oxygen mask. Now I know they were literally speaking the record into existence.
Voice writing has been recognized by the NCRA and is used in many jurisdictions. It's another pathway into the profession that doesn't require mastering the stenotype machine.
Let's circle back to the hobbyist angle, because Daniel specifically asked about it. The Open Steno Project, Plover, the custom keyboards — is this actually a viable way to learn?
The community has developed extensive learning resources. There's a free online textbook called "Learn Plover" by Zack Brown. There are typing tutors specifically designed for steno, like Typey Type. The Plover Discord server has thousands of members who share dictionaries, help beginners, and organize speed-building challenges. Some people reach professional speeds using nothing but a mechanical keyboard and Plover.
The barrier to entry is essentially zero. You can try steno on the keyboard you already own.
The caveat is that not every keyboard supports n-key rollover. Most laptop keyboards don't. But a decent mechanical keyboard with NKRO costs maybe sixty dollars. Compare that to a five-thousand-dollar professional stenotype machine, and you can see why the open-source approach is democratizing access.
There's something appealing about the idea of learning a completely different input method. It's like learning a musical instrument that also happens to produce text.
That's actually a great analogy. Stenographers describe their work in musical terms — they talk about rhythm, about flow, about the physical sensation of a well-executed chord sequence. The machine has a distinct feel. Some stenographers use very light key switches, almost touch-sensitive, because they're pressing so many keys so rapidly that heavy switches would cause fatigue.
I want to ask about the phonetic aspect, because Daniel mentioned this and it's counterintuitive. Most people assume you'd want to type letters to produce words. But phonetic typing is faster.
Because English spelling is a disaster. We have forty-four phonemes and over a thousand ways to spell them. The word "through" has seven letters for three sounds. "Though" and "through" and "tough" and "cough" all have the same letter sequence and completely different pronunciations. Phonetic writing bypasses the spelling problem entirely. You write what you hear, not what the dictionary says.
The stenographer hears "though" and writes THO, hears "through" and writes THRU, hears "tough" and writes TUF. The translation software knows which English word corresponds to which phonetic stroke.
And because you're writing phonetically, you don't have to think about spelling at all during the capture phase. The cognitive load is lower. You're not mentally consulting a dictionary for every word. You're just encoding sound.
Which must be especially valuable in courtrooms where you might encounter words you've never heard before. A technical expert testifying about some obscure chemical compound — the stenographer hears the sounds and writes them phonetically, and the dictionary can be updated later with the correct spelling.
That's standard practice. Stenographers maintain what they call a "job dictionary" — a custom dictionary for each case or client, loaded with the proper names, technical terms, and unusual vocabulary that will come up. Before a trial, they'll get a list of witnesses, exhibits, and key terminology. They pre-program steno strokes for all of it.
That's a level of preparation I never would have associated with this job. It's not just showing up and typing fast. It's doing research, building a custom lexicon, understanding the subject matter enough to know what words are likely to appear.
The best court reporters are deeply knowledgeable about the cases they cover. They have to be, because context helps disambiguate. If a witness says something that could be transcribed two different ways, the stenographer's understanding of the case tells them which one is correct.
Daniel also mentioned the WikiLeaks diplomatic cables and how fascinating he found the analytical quality of diplomatic note-taking. There's an interesting contrast here. Diplomatic cables are interpretive — they summarize, they analyze, they add metadata. Court transcripts are the opposite. They're strictly verbatim, no interpretation, no summary, no commentary. The stenographer's job is to disappear, to be a perfectly transparent medium.
That's harder than it sounds. Human beings naturally want to make sense of what they hear. We automatically filter, summarize, and interpret. Stenographers have to suppress that instinct. If a witness says something grammatically incoherent, you transcribe the incoherence. If they use a double negative, you capture it exactly. The transcript is not a cleaned-up version. It's the raw linguistic reality of what was said.
Which makes the AI question even more complex. An ASR system doesn't have the instinct to interpret, but it also doesn't have the judgment to know when something doesn't make sense and should be flagged. It'll confidently transcribe nonsense.
There's a concept in the field called "verbatim with intelligence." It's the idea that a truly accurate transcript requires understanding context, not just matching sound waves to words. A human stenographer knows when a speaker misspeaks and corrects themselves, and captures both the error and the correction appropriately. An ASR system might produce a clean sentence that's grammatically correct but not what was actually said.
Let's talk about the international picture. Daniel is in Jerusalem, he studied in Ireland. What does court reporting look like outside the United States?
It varies enormously. The stenotype machine as we know it is primarily an American phenomenon, though it's also used in Canada, Australia, and some other common law jurisdictions. In the United Kingdom, court proceedings are typically audio recorded rather than stenographically transcribed. The official record is the audio. Transcripts are produced later by transcriptionists who listen to the recordings.
That feels like a fundamentally different philosophy about what the record is.
In the US system, the stenographer's transcript is often considered the official record, and the audio is secondary or not preserved at all. In the UK, the audio is the official record, and the transcript is derivative. Both systems have their advocates. The UK approach avoids the stenographer shortage problem entirely but introduces its own issues — audio recordings can be unclear, equipment can fail, and producing transcripts for appeal can take weeks or months.
In Israel, where Daniel is, the court system uses a mix of audio recording and live transcription for certain proceedings. The Hebrew language presents its own challenges for both stenography and ASR — it's a Semitic language with a non-Latin script, right-to-left writing, and a complex system of vowel markings that are usually omitted in writing.
Stenotype machines have been adapted for many languages, but it's not universal. Japanese court reporting uses a completely different system. German courts typically use audio recording with subsequent transcription. France uses audio recording. The stenotype is most deeply embedded in the American legal system, which is partly historical accident and partly the result of the US system's unique demands — lengthy depositions, complex civil litigation, and a tradition of real-time access to the record.
I want to return to something Daniel asked about: the value chain of AI in this space. He said he thinks the real value sits on the other side — on taking human-generated meeting digests and making them useful across teams. Does that apply to court reporting?
In an interesting way, yes. The raw transcript is one thing. But courts and law firms are increasingly interested in what you can do with that transcript after it's produced. AI-powered legal analytics can process thousands of transcripts to identify patterns in judicial decision-making, analyze witness credibility, extract key facts, and summarize cases. The transcript becomes a dataset. That's where AI adds enormous value — not replacing the stenographer, but mining the output.
The stenographer creates the clean, verified, structured data, and the AI does the higher-level analysis. That's a division of labor that makes sense.
It's consistent with what we've seen in other domains. AI is extremely good at pattern recognition across large datasets. It's less reliable at the precision task of producing a single flawless artifact. The stenographer produces the artifact. The AI analyzes the archive.
There's a broader point here about professions that seem obsolete but aren't. I think a lot of people, Daniel included, assumed stenographers were just fast typists who would inevitably be replaced by better microphones and speech recognition. But the reality is much more nuanced. The profession has already adapted to technology — computer-aided transcription, real-time feeds, digital recording — without losing its core value proposition: a human brain guaranteeing accuracy in real time.
The shortage I mentioned is creating a strange dynamic where the profession is both threatened and in high demand. Stenographers who are in the field can command premium rates. Freelance deposition reporters in major markets can earn a hundred twenty, a hundred fifty thousand dollars a year or more. It's a skilled trade with a significant barrier to entry and a reliable market.
That's the kind of career that guidance counselors should probably be talking about more. Everyone's pushing coding bootcamps, but nobody mentions that you can make six figures capturing the spoken word with a specialized chorded keyboard.
The Open Steno community is trying to change that. They run outreach programs, workshops, online tutorials. The message is: this is a learnable skill, the tools are free, and there's a career waiting on the other side if you're willing to put in the practice hours.
I'm curious about the physical toll. We talked about repetitive strain injuries being lower with steno than with QWERTY typing, but there must be other physical demands.
Sitting in one position for hours, maintaining intense concentration, the mental fatigue. Many stenographers develop back and neck issues from the posture required. There's also a phenomenon called "steno flu" — not an actual illness, but a kind of mental exhaustion that sets in after a full day of real-time transcription. It's cognitively draining in a way that's hard to describe.
I believe it. Simultaneous translation is considered one of the most mentally demanding tasks a human can perform, and stenography is essentially simultaneous translation from speech to text.
With the added pressure that the output is a legal document. Interpreters at the United Nations work in teams and switch off every twenty to thirty minutes. Stenographers often work solo for hours.
Daniel asked how long it takes to train. Two to four years to professional certification, you said. What does the learning curve actually look like?
The first phase is theory — learning the steno alphabet, the chord combinations, the principles of phonetic writing. That usually takes three to six months. Then comes speed building, which is the long grind. You start at forty words per minute and gradually work up. Each speed level — sixty, eighty, a hundred, a hundred twenty, a hundred forty — requires hundreds of hours of practice. The jump from two hundred to two twenty-five, which is the certification threshold, can take six months by itself.
It's like training for a marathon, but for your fingers and your brain simultaneously.
There are plateaus. Almost everyone hits a wall at some point — a speed they can't seem to break through. Pushing past that requires deliberate practice, often with a coach or a structured program. The dropout rate reflects how many people hit that wall and decide it's not worth climbing.
For the hobbyist using Plover and a mechanical keyboard, the goals are different. They're not trying to get certified. They just want to type faster for their own purposes. What kind of speeds can a dedicated hobbyist achieve?
There are Plover users who have reached a hundred fifty, a hundred eighty words per minute. Some have gone on to get certified and enter the profession. Most are happy at a hundred to a hundred twenty, which is already significantly faster than the average QWERTY typist and fast enough for personal writing, coding, and note-taking.
I have to ask: have you ever tried it?
I have, actually. I downloaded Plover, watched a few tutorials, and spent a weekend learning the basic chords. I got to the point where I could write simple sentences at about twenty words per minute. It was humbling. Twenty words per minute. I type at ninety on QWERTY. But the sensation of chording — pressing multiple keys and having a whole word appear — was thrilling. I can see why people get hooked.
I'm picturing you hunched over a keyboard, muttering phonetic combinations to yourself, completely in the zone.
It's meditative, in a strange way. You have to let go of thinking about individual letters and trust the chords. When it flows, it feels like playing music. When it doesn't, it feels like your brain has been replaced with a bag of wet sand.
That's the hobbyist experience. Imagine doing it for eight hours in a courtroom with someone's liberty at stake.
That's why I have enormous respect for the profession. It's a specialized skill that serves a fundamental civic function. The record of what happens in court is the foundation of the entire appellate system. Without an accurate transcript, there's no meaningful appeal. Due process literally depends on these people getting every word right.
That's a good place to start wrapping up. The technology is fascinating, the hobbyist community is delightful, but the core of this is a profession that quietly upholds a pillar of the justice system.
It's adapting. I think Daniel's question about AI is the right one to ask, but the answer is more interesting than "AI replaces stenographers." The likely future is a hybrid: AI-assisted stenography, where the machine suggests and the human verifies, or stenographers transitioning partially into editing roles for certain types of proceedings while remaining essential for high-stakes live work. The technology doesn't eliminate the need for human judgment. It changes where that judgment is applied.
One last thing: I want to note that Daniel asked about buying a stenotype machine. The answer is yes, you can buy one new for a few thousand dollars, used for less, or you can build a hobbyist setup for under two hundred dollars with Plover and a compatible keyboard. The community is out there. The resources are free. If you want to become the person in the coffee shop with the weirdest input device, the door is wide open.
And now: Hilbert's daily fun fact.
Hilbert: The Greenland shark can live for over four hundred years, making it the longest-living vertebrate known to science. Researchers determine their age by radiocarbon dating the eye lens nuclei, which are formed before birth and never regenerate.
Four hundred years. That shark was swimming around when Shakespeare was writing plays.
That's unsettling and I'm not sure why.
This has been My Weird Prompts. Our producer is Hilbert Flumingtop, and we'd love it if you'd take a moment to rate and review the show wherever you listen. It helps more than you'd think.
We're back next time with whatever Daniel sends us. Should be interesting.
