The Pepys blog
Original research and notes on transcription – what people actually struggle with, and the honest fixes.
How Many Hours of Audio Does a Typical Qualitative Study Produce?
Interview studies run about 9-17 hours of audio; focus-group studies about 4-12. Here is the research behind those totals.
Read →How Long Is a Research Interview? Length Benchmarks by Type
Duration benchmarks by interview type, from 20-minute surveys to multi-session oral histories, plus what each length means for transcription time.
Read →How Long Is the Average Deposition, and How Many Transcript Pages Does It Produce?
No dataset sets a true average. Here's what the rules fix instead: a 7-hour federal cap, state limits from 3 to 20 hours, and about one transcript page per minute.
Read →How Much Audio Does a Podcast Produce in a Year?
A weekly show at the most common episode length makes 17 to 35 hours of raw audio a year. The full math, by episode length and publishing cadence.
Read →How Long Does It Take to Transcribe a Research Study by Hand?
A full qualitative study runs 120 to 210 hours of hand-transcription. Here's the math, from interview count to working weeks.
Read →How long is the average recording?
There is no single number – there's a distribution. Sorted by what's being said, published data on podcasts, meetings, interviews, lectures, sermons, sales calls and depositions.
Read →Subtitle reading speed standards: the numbers Netflix, the BBC, and WCAG actually use
Every serious caption style guide draws a line for how fast text can move past a viewer. Here are the real ceilings, in words per minute and characters per second, and why they sit below how fast you read a book.
Read →How many filler words per minute is normal?
Um, uh, false starts, repeats – linguists have actually counted them. Here is the per-minute number, and why it is a range rather than a constant.
Read →Reading speed vs listening speed: the research on why you read faster than anyone can talk
The numbers are close to settled science. Silent reading beats speech by a wide margin, and unlike audio, you set the pace. Here's what the studies actually found.
Read →How big is the transcription market in 2026? The numbers, by scope.
There is no single "the transcription market is $X" figure. There are several, measuring different things, growing at very different speeds. The gap between them is the whole story.
Read →Does audio quality affect transcription accuracy? Yes – and here's how much, by the numbers.
A tour of the published ASR research – SNR, noise, reverberation, mic distance, sample rate, and compression – and how each one moves word error rate before the model gets a say.
Read →How big is an hour of audio?
The honest answer is a range, not a number – from about 58 MB to 600 MB depending on format. Here's the math behind every one, and the 25 MB wall it keeps running into.
Read →Transcription in academic research: the hours and dollars nobody budgets for
A qualitative thesis is a burst of interviews, then quiet. Here is how much audio that actually is, what it costs to turn into text, and why the tool has to answer to your ethics board.
Read →Transcription accuracy by domain: why the same AI model breaks on specialized speech
General speech recognition is close to solved on clean read audio. The moment jargon, drug names and case citations enter, the same model's error rate can jump two to ten times. Here is the spread, by domain, with the data.
Read →The cost of not transcribing meetings
The average knowledge worker gives up about 18 hours a week to meetings and calls a third of them pointless. The bigger cost is the one no one bills you for – the detail that's gone by morning.
Read →Dictation vs typing speed: the case for talking first
Speaking runs near 150 words a minute. Typing averages 40 to 52. The gap has been measured directly, and practice barely closes it.
Read →The real AI transcription limits in 2026: every major tool's hard ceiling
The AI you already pay for is not a transcription tool, and its limits prove it. Here is the exact number for ChatGPT, Claude, Gemini, NotebookLM and Otter - verified against vendor docs.
Read →How accurate is AI transcription? The benchmark numbers, read honestly.
Published speech-recognition benchmarks give a clear answer - and a badly misleading one, if you read only the headline. The honest version depends entirely on the audio.
Read →How many words are in an hour of audio?
Roughly 8,000 to 9,000 - but the real number depends on who is speaking. A walk up the speaking-rate ladder, from Churchill's cadence to an auctioneer's, and what an hour of talk costs you in pages, reading time, and dollars.
Read →Transcription costs in 2026: every published rate on one axis
From a third of a cent a minute to over eight dollars a page. We put every publicly listed 2026 rate on one line – dollars per audio minute – and explain why each tier costs what it does.
Read →Which languages AI transcription struggles with – and why
Accuracy isn't one number. On published benchmarks it swings from single digits to triple digits depending on the language, the dialect, and whether the speaker switches languages mid-sentence.
Read →How well does AI handle crosstalk? The honest answer, with the data.
The counterintuitive part: AI is good at telling voices apart. It is bad at hearing two people at the same time – and that gap is exactly where transcripts fall apart.
Read →How long does it take to transcribe an hour of audio?
The 4:1 rule of thumb, the honest spread from a fast pro to a verbatim slog, and why AI collapses hours of work into seconds. A numbers-first timeline for planning a project.
Read →The State of Transcription in 2026
AI transcription is genuinely near-solved on clean, high-resource English – and genuinely uneven everywhere else. A data report on where the technology actually stands.
Read →We read 100 Reddit threads about transcription. Here's what everyone actually hates.
Not keyword research – real people, venting in real threads, about tools that failed them mid-project. Four complaints came up again and again.
Read →
Keep Pepys in your Google results
If our transcription guides are useful, add Pepys as a Preferred Source in Google.
