SpeechCraft Studio
When I was working at ShazaCin Accessible Media on the audio description app describeAT, I was already deep into sound editing and developing a new interest in AI. What I didn’t expect was how those threads would eventually weave together into something entirely different.
SpeechCraft Studio began as an accident — or rather, as an experiment that got out of hand in the best possible way.
The Burrow
I thought I’d build something simple to help me clean out loud breaths from audio. It’s a time-consuming job, and I figured AI could automate part of it. After watching the first breath-smoothing version work, I realised the possibilities were much bigger than I’d imagined. I just had to say “I wonder if I could …” and keep going.
SpeechCraft Studio is a bit like the Burrow from Harry Potter — one room built on top of another, each addition sparking the next.
The rooms in the house
Room one: breath smoothing. The original idea. RMS-based detection that finds breaths and attenuates them automatically.
Room two: transcription. Once I had the basics of an editor, I wanted something like Descript — where you edit audio the way you’d edit text. If I said “um,” I could just delete the word. So I added transcription.
Room three: line placement. If there’s a transcript, I thought, can’t I automate where audio description lines go? Anyone who’s placed AD lines knows the hours of work it takes to get them in the right place. So I built a line placer system — import your script file, and it matches against the transcript to position the lines automatically.
Room four: studio recording. From there came the studio recorder. Two people — the voice artist and the director — see the same script. As recording happens, SpeechCraft Studio places the lines in real time. You can say “oops” or hit redo if you make a mistake.
I also built in Braille display support so the next line pops up automatically, and a way to connect two monitors (or a second laptop over the network) so both director and voice artist can see what they’re reading.
The additions kept coming. Text-to-speech for when you want an AI voice at any stage. Effects — compressor, de-esser, EQ, noise gate, normaliser, room tone match — because I don’t always have the budget for expensive audio plugins. A bug reporter that pre-fills GitHub issues for you. And, of course, all of it designed to work with NVDA, JAWS, and other screen readers.
The making of it
This project took months of work — contrary to the belief that you just say the word and an app appears. It took a lot of testing, learning, and retrying. Sometimes I got frustrated. Sometimes I broke something that previously worked. I ran into walls where I didn’t know how to make the app behave the way I wanted it to.
But despite those moments, I had tremendous fun. And the feeling I get when something finally works is the same happiness and pride I feel when I wear a sweater I’ve made myself.
In the months between starting SpeechCraft Studio and today, I’ve learned a lot about building with AI. Combined with my short time working as an apprentice with an experienced app developer at BGC, the way I work now is not the way I worked at the beginning. The journey itself was part of the reward.
Try it
SpeechCraft Studio is a free, open-source desktop audio editor for Windows. It’s natively accessible for screen reader users, and it’s available as a single .exe file — no installer needed.
I’d love to hear what you think. Send me any feature requests or let me know if you find a bug. And if you’d like to contribute, the repo is open to pull requests.
I hope you enjoy making audio.
— Tracy