Writing on voice AI, benchmarking, and building in the open.
15 posts so far: the benchmark, the methods, and the stances behind the product. More land as the research does.
How speech to text actually works (the words, not the captions)
A transcript is a string of words. Captions and meeting minutes are other products.
- ResearchAug 19, 2026
How AI sound effects actually work (the bed, not the door slam)
A bed sits under speech. A hit punctuates a picture. Most generators blur the words.
Read · 6 minRead post → - ResearchAug 12, 2026
What neural TTS actually is (the pipeline, not the magic mouth)
Neural TTS sounds human until a name, a number, or a stage direction proves it is still a machine with stages.
Read · 7 minRead post → - ResearchAug 5, 2026
How AI dubbing actually works (when you refuse the magic button)
Dubbing is a pipeline with review points, not a black box that 'just translates the video.'
Read · 7 minRead post → - ResearchJul 29, 2026
Pick the engine for the job, not the demo
A single engine is a demo. A routing table is how you ship without burning the wrong pass on every line.
Read · 5 minRead post → - ResearchJul 22, 2026
How to make an AI audiobook without one engine doing every job
A multi-engine audiobook workflow is not a stunt. It is how you stop paying final-pass rates for every draft.
Read · 7 minRead post → - ResearchJul 21, 2026
Your engine ignores stage directions (unless it doesn’t)
Emotional text to speech is not a filter you slap on at the end. It is direction in the script, and only some engines will follow it.
Read · 6 minRead post → - ResearchJun 27, 2026
Lawful, not lawless: the industries every other AI voice tool turns away
We do not deplatform lawful businesses by industry or viewpoint. Who refuses firearms, cannabis, crypto, and gambling, with the receipts, and the voice tooling they are pushed to need.
Read · 24 minRead post → - ProductJun 16, 2026
We built voice cloning. Then we didn't ship it.
The feature is built and flag-gated off. Doing it honestly is harder than building it, so we did the building and waited.
Read · 4 minRead post → - ResearchJun 16, 2026
AI audio has to say it's AI now. Here's the law.
Labeling AI audio is shifting from good manners to law. Here is the map, and our position.
Read · 4 minRead post → - ProductJun 11, 2026
Why our free tier has an unlimited engine
Open weights changed the arithmetic of a draft. We priced the free plan on the arithmetic.
Read · 3 minRead post → - EngineeringJun 11, 2026
How we verify every fixed sample on this site
Engines intermittently truncate long takes. Every fixed clip here is transcribed back and checked before it ships.
Read · 3 minRead post → - ResearchJun 10, 2026
The open voice benchmark, and why we run it
Vendor demos tell you which engine sounds best in the vendor's hands. We wanted numbers nobody here can tilt.
Read · 3 minRead post → - EngineeringJun 10, 2026
How we measure voice latency (and why vendors won't)
Wall-clock to the complete audio file, median of three, same script, same gateway. The script is in the repo.
Read · 3 minRead post → - ProductJun 10, 2026
You own what you make here
No meter renting you your own audio, no watermark, no clause that quietly keeps your work.
Read · 3 minRead post →
Every number in these posts is interpolated from the same source files the product reads, so the writing cannot drift from the site.
Want to hear the work behind the writing?
The console is live. Generate audio from any engine and see the open benchmark.