I’m Leif and I wanted to share the new product I’ve been working on recently, TurboScribe (https://turboscribe.ai).

It’s pretty simple: unlimited Whisper transcription (starting at $10 per month). It supports large-v2, small, and base models. And yeah, it really is unlimited. The most active users have transcribed 1k+ hours using Whisper large-v2 (in other words, you can transcribe all 30 x 24 x 60 = 43200 minutes of your life every month if you want!).

When I started building this a few months ago, I got connected with some initial users with higher volume transcription needs (i.e. hundreds/thousands of hours) and offered it for free. This helped me build out the transcription engine and iterate on it until it was beefy enough / efficient enough / fast enough to handle that level of load (they were also really impressed with the quality of Whisper transcription – so was I when I first started toying with it earlier this year!). Initially I wasn’t sure if I could get things efficient enough to be able to offer an “unlimited” service (especially one that could support people with higher-volume transcription needs), but it turns out that this actually is possible, which was cool.

Long story short: I’ve been really pleased with how it’s gone so far and so I’m excited to start sharing TurboScribe more broadly now. It’s transcribed over 17k hours so far (and less than half of that is my own personal testing :P), which I guess isn’t so big in the grand scheme of the internet, but I’m excited to see where this goes.

On a personal note, this has been the most fun I’ve had working on something over the past couple years and I’m excited to continue shipping more. Some things in the queue: improved file/folder management, transcript editing tools, and an API.

Tech details: It’s all Clojure/ClojureScript, with a Postgres database underneath. HTMX (I could write many love songs about HTMX) made the more dynamic UIs a breeze to build (live transcription progress, lazy loading, multi-step forms). Another Clojure program manages the transcription engine (heavy use of Clojure core.async) to allocate transcription jobs to a pool of GPU machines running Whisper, retry in the event of occasional failures, etc. Cloudflare R2 has been great, especially when it comes to handling large uploads – it’s awesome to be able to transfer large files to a transcription machine without huge data transfer costs. I could go on but not ashamed to say I’m building on the shoulders of giants. :)

There’s a free tier (4 uploads daily, up to 30 minutes each) and I would love for y’all to try it out. Happy to answer any questions (emails also welcome [email protected]) and otherwise thanks for checking it out!