Nikke Scribe
Transcribe 10-hour WAV files. No subscription required.
5 followers
Transcribe 10-hour WAV files. No subscription required.
5 followers
Why pay for a monthly subscription just to transcribe a few recordings? Most transcription services lock you into recurring plans, even when you only need them occasionally. Nikke Scribe offers a simpler alternative. Transcribe up to 10 hours of audio or video in a single file, with AI-powered transcription delivered in seconds. Pay only when you need it. No subscriptions, no recurring fees, no wasted credits. Built for interviews, meetings, podcasts, lectures, and other long recordings.
This is the 2nd launch from Nikke Scribe. View more
Nikke Scribe
Launching today
Why pay for a monthly subscription just to transcribe a few recordings?
Most transcription services lock you into recurring plans, even when you only need them occasionally. Nikke Scribe offers a simpler alternative. Transcribe up to 10 hours of audio or video in a single file, with AI-powered transcription delivered in seconds.
Pay only when you need it. No subscriptions, no recurring fees, no wasted credits.
Built for interviews, meetings, podcasts, lectures, and other long recordings.
Free Options
Launch Team
LoomShow, don't just tell with Loom
Promoted
After Ph.D in CS, I started building this because I found it ridiculous that you have to sign up for a monthly subscription just to transcribe a couple of recordings. You might have a 3-hour interview or a 10-hour meeting to transcribe, but that doesn't mean you need a transcription service every month.
So I built Nikke Scribe. It handles long audio and video files, including WAV files up to 10 hours, and transcribes them in seconds. No subscription required. Just pay for what you use.
I'd love to hear what you think, especially if you've run into the same problem with transcription tools.
Why is Nikke Scribe several to tens of times faster than conventional transcription services?
We built a sandbox where Fable could study the Whisper model through hundreds of experiments, from reimplementing the FFT used in audio processing to discovering structural weaknesses in Whisper itself. Rather than simply using a frontier model to write code, we used it as a researcher to investigate and optimize speech recognition. The resulting optimizations, down to the FFT level, are what make Nikke Scribe so fast.
Ultra-fast transcription isn't just about saving time. It also means using less compute.
As semiconductor supply becomes increasingly constrained, companies need to make more efficient use of the chips they already have.
In a recent report,
@cerebras
explained that semiconductors are often replaced not because they wear out, but because newer chips become more economical to operate. In other words, replacing perfectly functional hardware can make more economic sense than keeping it.
But what if we could improve the economics of existing hardware through software alone?
That's the question we're exploring with Nikke Scribe: how far can we push computational efficiency by optimizing the software itself, rather than relying on the next generation of chips?
----------------
One answer is 100% utilization. Ideally, we should keep inference hardware running at a consistent load, 24 hours a day, until the next generation of chips replaces it.
Conventional transcription services take several minutes to transcribe an hour of audio, or at least around 30 seconds. Consider a simple case: a second user submits a request just one second after the first. If the system processes requests sequentially, that user may have to wait 30 seconds before processing even begins. Few users are willing to wait that long.
When there are no requests, the hardware sits idle. Its utilization drops, but the cost of operating it doesn't disappear. Infrastructure providers pass those costs on to transcription services, which in turn pass them on to customers through subscription pricing.
This is why transcription speed matters beyond the user experience. Faster inference changes the economics of compute itself.
----------------
Our goal was to bring that waiting time down to 10 seconds. We've now confirmed that, in our fastest observed case, Nikke Scribe transcribed one hour of audio in just 7 seconds.
At this speed, users can get their transcripts without leaving their desks to wait. At the same time, faster processing means each request occupies the server for less time, allowing us to serve more users with the same hardware and offer transcription at a lower price.
The less time each audio file occupies our computing resources, the lower the overall cost of transcription becomes. That's the economics we're trying to change with Nikke Scribe.