The words as spoken
No summarising, no tidying, no model deciding your subject meant something more polished. What was said is what lands in the file — which is the only version worth quoting from, and the only one worth checking against the audio.
Interview transcription
Your interview, transcribed where you recorded it.
Drop the recording in and get the words back with timecodes — your subject’s name spelled right, nothing summarised, nothing sent to a transcription service and nothing charged by the minute.
macOS 14.4 or later · Apple silicon · $29 once, not per minute · 7-day trial
How it works
From tape to transcript in three steps.
No account, no upload, no waiting behind other people’s jobs in someone else’s queue.
- 01
Bring the recording over
AirDrop it from the phone or recorder you interviewed with, or drop the file straight in — M4A, MP3, WAV, or the video if you filmed it. A folder your recorder syncs into can be watched, so a week of fieldwork queues itself.
- 02
Teach it the names first
Add your subject’s name, the company, the drug, the village, the acronym everyone in that field uses. Import a list you already keep as CSV. Every transcript after that spells them your way instead of guessing phonetically.
- 03
Work from the transcript
Every line carries a timecode, so a quote leads back to the second it was said and you can check the tape before you print it. Export TXT for the write-up, SRT if the interview is going on camera.
What matters in an interview transcript
Verbatim, checkable, and spelled correctly.
Three things decide whether a transcript is usable: whether it says what was said, whether you can get back to the audio, and whether the names are right.
Names spelled right, every time
Generic models fail hardest on exactly what an interview is full of: surnames, place names, company names, jargon. Add each one to the vocabulary once and it is fixed everywhere afterwards — dictation, files and the agent’s calls alike.
A timecode on every line
Quotes have to be checkable. Timestamps let you jump back to the moment before you commit a line to print, and they survive export, so a fact-checker or a supervisor can do the same.
An hour of tape, unmetered
Transcription services charge by the minute and typically want a dollar or more for each one. A ninety-minute interview costs nothing here beyond the licence, and so does the fifteenth one this month.
A folder that keeps up with fieldwork
Point Kekoso at the folder your recorder or phone syncs into and each new file is queued as it arrives. Twenty interviews from a trip become twenty documents without twenty drags.
Interviews that are not in English
Parakeet covers 25 European languages, Whisper reaches the long tail of 99, SenseVoice handles Chinese, Cantonese, Japanese and Korean. The recognition language is set per model, or left on Auto.
Where the recording goes
You promised them it stays with you.
Every interview starts with some version of that sentence — a confidentiality clause, an ethics approval, a source who agreed to talk on condition, or just a person trusting you with something. An interview transcription service asks you to upload the recording as step one.
Kekoso reads the file where it lies. The pipeline makes no network calls, nothing is copied anywhere, and the transcript is a plain local file you can archive or delete. Whatever you promised the person on the tape stays true without you having to read anyone’s terms of service.
How to verify that yourselfWhat it does not do
The limits, up front.
It does not label who is speaking
An interview recorded on one microphone comes back as one continuous transcript with timecodes, not split into interviewer and subject. Kekoso does not do speaker diarization. In practice, marking up two voices in a transcript you are going to read closely anyway is quick; if you need labels done for you, this is the wrong tool and it is better to know now.
The microphone decides the result
A recorder in the middle of a café table, two people talking over each other, an air conditioner behind the voice — no model recovers those cleanly, and a bigger model will not fix what the microphone never captured. Close, quiet, one voice at a time is what makes a transcript usable.
You still read it
On-device recognition is good, not perfect, and proper nouns are where every model struggles most. Plan a pass with the audio open, especially before quoting. The vocabulary removes most of the repeated corrections, not all of them.
Apple silicon only
macOS 14.4 or later · Apple silicon. Recognition leans on the Neural Engine, which Intel Macs do not have.
Questions
Before the next interview.
- How do I get a transcript from interview recordings I already have?
- Drop the recording into Kekoso and let it run — the route to a transcript for an interview is the same whether it is one file or thirty. Recognition happens on your own Mac and produces a transcript with timecodes, which you can copy out or export as TXT, SRT or VTT. There is no upload, no queue and no per-minute charge, so a ninety-minute interview costs the same as a two-minute one.
- How does this compare with an interview transcription service?
- A service charges per minute of audio and a human or a model does the work on their machines, which means sending them the recording. Kekoso is a one-time purchase and the recording stays where it is. What a good service still gives you that this does not is speaker labels and a human proofreading pass — if your work depends on either, factor that in.
- Will it separate the interviewer from the subject?
- No. Kekoso does not do speaker diarization, so a two-person interview on one microphone comes back as a single transcript with timecodes. Calls recorded on the Mac itself are the exception: your microphone and the other side are captured as separate tracks and stay labelled.
- Can it handle names and technical terms?
- That is what the vocabulary is for. Add the surnames, place names, company names and jargon your material is full of — one at a time, or imported from a CSV list you already keep — and every transcript afterwards writes them your way rather than guessing phonetically.
- Is the recording uploaded anywhere?
- No. The speech models run on your Mac and the transcription pipeline makes no network calls at all. The file is never copied and the transcript is a plain local file. For interviews conducted under a confidentiality agreement or an ethics approval, that difference is usually the whole question.
- How long does an hour of audio take?
- Less than an hour, and how much less depends on the model you pick. Each model card shows its measured speed and word error rate on the same benchmark, so you can trade accuracy for time deliberately rather than guessing. Nothing is queued behind other people’s jobs, because there is no queue.
- What formats can I export?
- Plain text for the write-up, SRT and VTT if the interview is going on camera. All three carry the timecodes, and TXT is there when you want just the words.
- What does it cost?
- $29 once, for up to 3 of your own Macs, with a 7-day trial and no card to start. Interview transcription is one part of the app — dictation, call recording, subtitles and the local MCP server come with it.
Pricing
Buy it once. It's a tool, not a tenancy.
Four jobs that usually arrive as four subscriptions, running on hardware you already paid for.
$29once
7-day free trial. No card to start.
- Dictation, files and links, call recording and the MCP server — all of it
- No per-minute billing, no monthly cap, no usage dashboard
- Use it on up to 3 of your own Macs
- Free updates through the whole 1.x line
- Works offline indefinitely; no license check to phone home weekly
- 30-day refund, no argument, no questionnaire
macOS 14.4 or later · Apple silicon · payment handled by Lemon Squeezy
Try it on your own words.
7 days, every feature unlocked, no card and no account. If it does not earn its place in your week, delete it and nothing follows you.
macOS 14.4 or later · Apple silicon