Somebody in Clearwater has eleven hours of their father on a hard drive and hasn’t opened the folder in fourteen months.
I’ve met several versions of that person. They did the hard thing. They sat down with a parent, asked the questions, kept the recorder running through the awkward parts. Some of them did it in the last year the parent was well enough.
Then the file landed on a laptop and the project stopped, because the next step has no obvious first move.
How many words is eleven hours of audio?
Roughly a hundred thousand words of speech. About four hundred pages if you printed the transcript.
Which sounds like a book and is nothing like one. Speech is repetitive, circular, and full of half-finished thoughts that the speaker abandoned mid-sentence because they remembered something better. It doubles back. It contradicts itself. It contains long passages about a neighbor whose name means nothing to anybody.
A finished memoir runs sixty to ninety thousand words. So the arithmetic looks like it works. That’s the trap. You’re not trimming ten percent off the top. You’re keeping maybe ten percent and rebuilding the rest.
The ten percent you keep is unavailable anywhere else, in any form, ever again.
Why do memoir projects stall after the interviews?
Because the next step produces nothing you can look at.
Recording felt like progress. You could see the file size. Writing will feel like progress too, once it starts. The step in between is reading a hundred thousand words of somebody talking while making notes about structure, and at the end of two full days you have a document that looks like a list.
Nobody sustains that on enthusiasm. It’s to be a process you follow when you don’t feel like it.
The other reason people stall is that opening the file means hearing the voice. If the person has died since the recording, that’s not a small thing and it’s worth naming. Several people have told me they couldn’t do the transcription themselves for exactly that reason, and having somebody else handle it was the thing that let the project continue.
Step one: transcribe it, badly
Use automated transcription and accept that it’ll get names wrong.
Perfect transcription is a trap at this stage. You’re producing a searchable document, not a publishable one. Machine transcription of clear audio is good enough to find things in. That’s the entire requirement right now.
What matters is the timestamps. Keep them. Every time you find something usable in the transcript you’ll want to go back to the audio and hear how it was said, and without timestamps you’re hunting through eleven hours.
Fix names and places as you go, since those are what you’ll search on. Leave everything else wrong.
Back the whole thing up in two places, one of them not in your house. I’ve written about the recording side of this and the same rule applies here, doubly, because now you have the only copy of both the audio and the transcript.
Step two: the index nobody wants to make
Read the transcript once, start to finish, and build one document as you go.
For each passage worth keeping, write a single line: what it’s about, roughly when in the subject’s life it happened, and the timestamp. Not a summary. A pointer.
You’ll end up with two or three hundred lines. That document is the most valuable thing you’ll produce in this whole project, and it’s the reason to do this step properly instead of trying to write from the transcript directly.
Mark three things as you read. Anything that made you stop, because that reaction is data. Anything that contradicts something said elsewhere, because contradictions are where the real story sits. And anything they said only once, quickly, and moved past, because that’s usually the thing they weren’t sure they wanted to say.
Then sort the index by period. You’ll immediately see where the coverage is thick and where there’s a decade with four lines in it.
How should a memoir be structured?
Chronology isn’t a structure. It’s a default, and it produces the document nobody in the family reads twice.
A book needs one question it’s answering. Not a theme. A question.
Read your index and ask what this person’s life was really about. Sometimes it’s obvious. A man who left one country and spent fifty years in another is answering a question about belonging whether he framed it that way or not. Sometimes it takes three passes.
The test is whether you can say it in one sentence without using the words journey or legacy. If you cannot, keep reading the index.
Once you have the question, most of the cutting decides itself. Material that serves the question stays. Material that doesn’t goes into an appendix or a separate family document, and I’ve written about when that second document is the right deliverable.
Step four: the gaps you now have to fill
Your index will show you what you failed to ask. Everybody’s does.
If the person is still alive, that’s a second interview session with a specific list, and it’ll be the most productive session of the project because you now know exactly what’s missing.
If they’re not, the gaps get filled from documents. Census records for household composition. Obituaries for relationships nobody mentioned. Land records, city directories, newspaper archives. The research side of this in Tampa Bay covers where all of that lives, and the short version is that a great deal of it is free on a fourth floor in downtown Tampa.
Other people fill gaps too. Siblings, cousins, the friend who worked with them for twenty years. Somebody who knew this person professionally has a version of them their children have never heard.
Step five: write, and stop transcribing
This is where most amateur attempts go wrong and the failure has a signature.
The draft reads like a transcript with punctuation. Every digression preserved. Every repetition intact. Every name mentioned. It’s faithful and unreadable, and the writer knows something is wrong but reads the faithfulness as a virtue.
Fidelity to the recording isn’t the goal. Fidelity to the person is.
Those come apart constantly. A man tells a story in the recording across four separate passages, twenty minutes apart, doubling back twice. On the page it becomes one scene, in order, at a third of the length, and it sounds more like him than the transcript does.
Keep the exact words where the exact words matter. His phrase for a thing. The way he described his mother. The sentence he said twice without noticing. Those go in verbatim and everything around them gets built.
Should a memoir be in the subject’s voice or the writer’s?
A decision you have to make early and then hold.
First person, in his voice, is the strongest version and the hardest to write. Everything has to sound like him, so you’re doing the job I get paid for, and the failure mode is a book where he sounds like a man giving a speech.
Third person, in your voice, is easier and creates distance. It also lets you say things he wouldn’t say about himself, and for some subjects that’s the only way the book works.
Mixed, with his recorded words set apart from your narration, is the most forgiving. It’s also honest about what it is, and for family archives it reads best, because the reader can hear him directly and then have context supplied.
Pick one before you draft. Switching at chapter four means rewriting chapters one through three.
How long does it take to write a memoir?
Honest numbers, because the fantasy version is what kills projects.
Transcription review and correction on eleven hours: two to three full days.
The index: two days if you do it properly, and you’ll want to stop after four hours on the first day.
Finding the spine: a week of thinking, most of it not at a desk.
Drafting sixty to ninety thousand words: six to twelve months at an amateur pace, working evenings and weekends, assuming you don’t stop. Most people stop.
That’s the real reason people hire somebody. Not because the work is mysterious. Because eleven hours of audio turns into roughly four hundred hours of work, and they have a job.
When should you hand the recordings to a professional?
Three situations where doing it yourself is the wrong call.
The subject is unwell and the clock is real. A project that takes you three years and a professional eight months isn’t a comparison of cost, it’s a comparison of whether the book exists while they can read it.
You cannot listen to the audio. That’s a legitimate reason and not a weakness. The work can be done by somebody who didn’t love this person, and sometimes that’s precisely why it gets done.
You’ve tried and stalled twice. Two stalls is information. A third attempt with the same method produces a third stall.
What I do with eleven hours of audio is the same five steps above, done full time, with the difference being that I’ve done it 54+ times and I don’t get to stop when it’s boring. I work with people across Pinellas County and I can sit at the table for the follow-up interviews. That matters more than it sounds once the gaps need filling.
What to do with the audio once the book exists
The recordings don’t become waste when the manuscript is done. They become the more durable half of the project.
A book is an edited argument. The audio is the evidence, and a grandchild forty years from now will care about the voice more than the prose. Keep both and keep them separately.
Practical arrangement: the finished book for reading, the full transcript as a searchable document, and the audio files with their index, stored in two places with one outside the house.
Consider whether any of it belongs somewhere institutional. If the subject lived in this county and did something documented, a local archive may want the material, and several of them collect exactly this. The Dunedin museum runs an oral history initiative and the wider museum piece covers who else holds a record.
And label the files with real names and dates. A folder called Dad Interviews means nothing to the person who inherits the drive.
The mistake that costs the most
Waiting for the second round of interviews before starting the work on the first.
People do this constantly and the logic holds up until you try it. Why index eleven hours when there might be fifteen? So they wait for a visit, then a holiday, then a better week, and the audio sits.
The index is what tells you what the second session should cover. Doing it first makes the follow-up interview sharper and shorter. That matters when the subject tires after ninety minutes.
The same applies to the structure. Find the spine from what you have. It’ll change when new material arrives and it’ll change usefully, because you’ll know what the new material is for.
Work with what is on the drive now. That’s the only material anybody is guaranteed.
Can AI turn interview recordings into a book?
Partly, and the parts it does well aren’t the parts that are hard.
Transcription: yes, and you should use it. That’s a solved problem and paying a human to type is a waste.
Summarizing the transcript: adequate, and mildly useful for building the index faster. Check everything, because it’ll confidently attach the wrong name to the right story.
Finding the spine: no. A model will tell you the book is about resilience and family. It says that about every life. It cannot hear the one sentence somebody said quickly and moved past, because recognizing that requires knowing why a person would avoid it.
Writing in the subject’s voice: no, and this is where the failures are worst. A model produces competent prose that sounds like a man giving a speech. The specific rhythm of somebody’s speech is the entire asset and it gets averaged away.
Use it for the mechanical passes and do the judgment yourself. I use it that way every day and it’s never once told me what a book was about.
If you do nothing else
Get it transcribed and get the transcript backed up.
Audio degrades, formats go obsolete, and a hard drive in a Florida closet is on a clock. A searchable text file is small, durable, and readable by anything.
Even if the book never happens, a transcribed and indexed set of interviews is a real artifact. A grandchild can search it. A cousin can read it. Somebody in forty years can find the answer to a question nobody has thought to ask yet.
The book is the better outcome. The transcript is the one that stops the loss.