Skip to main content

Kajo Blog

Guides, comparisons,
and engineering notes.

How to transcribe, search, and question recordings without uploading them: honest comparisons with the tools you may use now, workflows for interviews, depositions, and sessions, and notes on how Kajo's on-device search and citations are measured.

Compare

3 posts

Use Cases

4 posts

Guides

13 posts
Guides

Arabic Transcription Software: MSA, Dialects, and Code-Switching

How Kajo Voice handles Modern Standard Arabic, Egyptian, Gulf, Levantine, and Moroccan dialects — plus Arabic-English and Arabic-French code-switching — with on-device transcription, speaker labels, and translation.

Guides

Code-Switching in Transcription: How AI Handles Multilingual Speech

What code-switching is, why it breaks most transcription tools, and how Kajo Voice's multilingual on-device transcription handles mid-sentence language transitions — honestly, including what to review.

Guides

Free Transcription Software in 2026: What "Free" Gets You When the Audio Can't Be Uploaded

A roundup of transcription tools that are free and keep audio on your machine — manual tools, the open-weight model from a terminal, local desktop apps, and Kajo Voice Free — with the cloud free tiers as the counter-example.

Guides

How to Transcribe an Interview Archive on Your Own Machine

A step-by-step guide to turning the recordings you already have into transcripts, speaker labels, summaries, and a private library you can ask with citations — on your own machine, with no upload. Formats, review, export, and what to expect.

Guides

Linux Desktop Transcription: Local AI Without a Mac Tax

Most voice AI apps skip Linux. Kajo Voice ships the same local-first app on Linux as on Mac and Windows — on-device transcription, speaker labels, a private library, and cited chat — as .deb, .rpm, and AppImage for x86_64 and ARM64.

Guides

Cited Chat Over Local Audio Archives: Ask Your Recordings, Keep the Receipts

How on-device cited chat turns a private transcript library into a searchable knowledge base — answers linked back to the source recording, scoped to the folder you are in, with no cloud upload.

Guides

Local vs Cloud Transcription: Which Is Right for You?

A complete breakdown of local AI transcription (Kajo Voice) vs cloud services (Otter, Fireflies, AssemblyAI). Privacy, cost, accuracy, offline use, and compliance — all compared honestly, with sourced prices.

Guides

Offline AI Transcription: Download Once, Then Disconnect

How Kajo Voice runs the whole loop — transcription, speaker labels, translation, summaries, cited chat — with the network off on Mac, Windows, or Linux: what downloads once, and what still talks to the network when it can.

Guides

Offline Transcription on Windows: Private, On-Device, and Searchable

How to transcribe audio and video privately on a Windows 10 or 11 PC — on-device, offline after setup, with speaker labels, a searchable library, and answers you can check. What it needs, how fast it is, and what never leaves the machine.

Guides

QDPX Export for NVivo Without Uploading Your Interviews

How qualitative researchers get REFI-QDA (QDPX) exports from the local-first audio knowledge base — ready for NVivo, ATLAS.ti, and MAXQDA — without sending audio to the cloud.

Guides

Transcription Accuracy: What to Expect from On-Device Transcription — and How to Check It on Your Own Audio

What published benchmarks say about the model Kajo Voice runs, what actually drives word error rate, how to get a better recording without a cloud cleanup step, and how to measure accuracy on your own audio before you pay.

Guides

The Privacy Guide to AI Transcription: What Happens to Your Audio

A detailed guide to what cloud transcription services do with your audio, the GDPR and confidentiality implications, and how to choose a transcription tool that protects sensitive conversations.

Guides

Whisper Large-v3 Turbo: The Model Kajo Voice Runs and Why

Whisper large-v3 turbo is OpenAI's speed-optimized Whisper variant — and Kajo Voice's single on-device transcription engine. Here's how it works, how it compares to the full large-v3, and why it's the right choice for on-device use.

Multilingual

3 posts

Engineering notes

4 posts
Engineering notes

What Reads This Number Next? A BM25 Defect That Dissolved

I filed a defect against my own search engine straight from the BM25 formula. Measured, the knob I proposed did not exist, the penalty was 25% rather than 40%, and within one script the length penalty cannot move the ranking at all — because fusion reads rank and throws the scores away.

Engineering notes

My Chunker Counted Characters and Called Them Tokens

The function deciding where every chunk of every transcript ended was a characters-per-token estimate nobody had compared with a tokenizer. It read 0.96 of the truth in Latin script and 0.48 in Devanagari — and the multilingual folder scoring 12 of 12 held ten documents of one chunk each.

Engineering notes

The Optimisation I Had Already Rejected, Twice

Forty-two percent of my answer time went on the model retyping text out of its own prompt, so I wired up the technique built for exactly that. It came out 44% slower. The payoff rides on one ratio — about 15% in my output — and I had already measured it, days earlier, into a flag I never read.

Engineering notes

When the Evaluation Is the Bug

I shipped a passage splitter behind 9,929 green tests, ran the evaluation with it on and with it off, and got identical numbers both times. Three readings from one harness: a change that never ran, a corpus that counted itself twice, and my fix for that, which collapsed 300 recordings onto 24.

Try it on your own recordings

Free to start, and nothing you import leaves your machine.