Scripty was born in December 2020/January 2021. Its core developer, Zero (they/them), deaf since birth,
was looking for a simple speech-to-text bot that actually respected their privacy. They couldn't find a single one at all.
Enter Scripty. It took around three months to get the first working version out... if you could call it working.
It was insanely CPU intensive and was generally poor in terms of everything.
In July 2021, Discord announced the introduction of the Message Intent. That's when v1 was abandoned.
Time passed. Scripty kept gaining servers despite being offline.
Eventually Zero's boyfriend pushed them a bit to maybe keep working on
Scripty since it was an awesome project. Despite life issues in general, in early February 2022, they figured they needed
something to work on. And so v2 was born.
v2 is a lot more efficient, using a different speech-to-text engine, streaming audio instead of buffering it,
among other optimizations that hugely reduce CPU usage. v1 was idling at about 80% CPU. v2 idles at less than 1%.
The difference in spikes between the two was even bigger: v1 spiked to 100% for multiple minutes when running the
algorithm, and v2 spiked to maybe 35% at worst for no longer than however long a user was talking.
Both are available on GitHub: you can find v1 here, and
v2 here.
However, through it all, Zero's core principles came first and foremost. Number one: don't compromise privacy.
v1 collected a lot of metrics about overall bot function, but never anything that could be traced back to a single server,
and only the absolute bare minimum was stored in its database. The same goes for v2. Number two: don't paywall things.
In both versions, core features are always free, forever. Premium versions of both exist, but neither locks core features.
Premium and free both have access to the same model and algorithm. Number three: don't be a dick.
Both of them always ask for consent before collecting anything that might be sensitive data,
and audio is absolutely never stored unless you opt in to it.
Along the way, there's been a lot of people and projects that helped out.
DeepSpeech by Mozilla: v1's underlying speech-to-text engine. An amazing project,
sadly abandoned due to the Mozilla Foundation laying off the employees who
worked on it during their 2020 budget cuts. GitHub repo
Coqui-STT: v2's underlying speech-to-text engine. Based off DeepSpeech.
Maintained by a German startup, and by the same people who originally
worked on DeepSpeech over at Mozilla. GitHub repo
v1's original testers.
I have no idea who you are at this point, but a huge thank you to all of you.
Even though it didn't work out, you did a lot to take Scripty to where it is today.