by Evan Balster
For all of us https://soul.dev/
John — thanks for the kind words. The probability method would fall under "AI" in the popular sense of that term, yes. So training with data is a consideration... but I'm wary of introducing biases by using data carelessly.
Matt — All our videos have shown imitone in genuine operation. In the original trailer, the actress was instructed to hum slowly and carefully because the 0.4 technology wasn't very reliable with fast music and open-mouth vocals. When user experience doesn't match our videos, it comes down to recording setup and singing skill. We're continuing to improve the technology so imitone can work well in poorer recording conditions and with less-experienced vocals — so you're pretty much spot-on.
Evan, I could not say it better than Jeff. This is my most favorite Kickstarter. (And I backed about 120)
It's a bit funny: I have to admit I tested the software just once sofar with mixed results. I am not very musical but I liked the idea. And I love your updates and the devotion you bring to it. Sometimes I thought I wouldn't hear again from your project and then an update came in that you improved the system.
If you would make a kickstarter for an Android App I would back it again.
The probability method sounds a bit like an AI approach to me. Is this correct? Could it be possible to train it with a bigger sound base like youtube? With cover songs of known music where the notes are available? Where the music was basically transcribed beforehand? Or would create it a mess because every artist sings the notes a bit differently or makes mistakes? Just thought AI could be the tool. But you probably need a good and big database.
Evan, you have my sympathies. Continue your good work as long as it takes and it feels good to you as well.
With due respect Evan, you sold the product on the premise of it working in the way that people imagine - and this was justified by people 'playing it' in the videos that were sported at the time. I remember being pretty impressed. If user experience doesn't match that, that's not the user's fault :)
Charlie, Ben, Robert, Jeff, Yehuda, Cam, Martin, MediaJolt: Thanks as always for the word of encouragement. You folks give me the confidence necessary to proceed with my insane plan. :)
Ben: A keen insight! This has actually been on my mind for a while. I an idea for a new workflow brewing, but I'm not ready to talk about it just yet.
Beh Kok Seng: imitone is absolutely coming to iOS and Android! Kickstarter backers will get access to these betas first. Machine learning "on the fly" is called "unsupervised learning". A lot is involved in doing it right, so it's probably out of scope for Stage 4. It's something we can come back to when we want to take imitone to the *next* next level. :)
Elliot: MIDI-CI will be a common framework for pretty much any useful two-way communication. It's up to manufacturers to implement those features, and to push for standardizing them. Be vocal with companies about the features you want! Or, if you make MIDI products, consider joining the MIDI Manufacturers Association yourself or via any MIDI-related organization you're part of. :)
Martin: My "tortoise" strategy is influenced in a big way by the stories I've heard about other, better-funded companies who have tried to pull off realtime pitch tracking. The R&D takes a long time. Bigger companies want to release products on predictable schedules, and smaller companies have to avoid running out of money. So most pitch trackers have gravitated to one of four routes that get the product to market faster while avoiding a huge investment in fully general pitch recognition:
1. Specialize with specific instruments (JamOrigin, Fishman Triple Play)
2. Specialize in manipulating audio (AutoTune, Melodyne, The Mouth)
3. Release it, market for Pro Audio, and hope for the best
4. Kill the project.
My company and my costs are tiny. imitone's sales are enough to cover those costs without being so high I need employees to help with customer support. This allows me to take my time and go all the way with the R&D.
I love it when you do a deep dive update!! Keep up the good work, Evan!!
I'm always so excited to hear about your progress. I think you've done - and are continuing to do - amazing work. Have you thought about having a mode where imitone processes the incoming data either after a couple second delay, or even after the entire recording is done? Essentially being able to use all the information to back-process, instead of needing to make split second decisions.
I always look forward to these updates. You're doing great work!
This is easily the best thing I've ever backed on KS. Evan - you amaze me with your continued devotion to this. I backed on a whim, and not being a musician or particularly interested in music (but fascinated by your idea) have not followed terribly closely - but am also excited to see one of your updates.
You are doing something amazing and all the hard work WILL pay off. Forget the naysayers or those that make demands of you. Continue on your path!
Great work and impressive devotion over the last several years.
Evan, I'm not a musician looking for a particular tool. I signed up for this project because I was fascinated by the possibilities and wanted to see where this could lead. You are clearly the right explorer for this job and like many here, I appreciate your thoughtful and considered updates. I hope the project continues to provide you with enough interest to keep going until you are satisfied with the result. I'm sure we will all be the beneficiaries of your hard work and deep insights. Thanks!
Nice one, Evan! Good luck at the show.
Why don’t you program imitone to including machine learning? Also converting it to a mobile platform where it is easy carry or use?
Please tell me they finally put state polling in MIDI 2.0. It causes so many problems not being able to ask instruments where there sliders and knobs are. In MIDI 1, you have to wait until the first time they move every time you connect things to get a state for them.
Great man, that you are still going and are still going strong. But why do you do everything on your own? Making imitone to know about scales and musical context requires yourself to know (so I guess) much mor about music then you already do. So for me it would make sense to partner up with a company like PG Music who sell Band In A Box. Perhaps their knowledge about styles would make the overall process faster. They will have all the music ready for imitone's future AI to train on.
Anyway, cheers to you.
This is fascinating! I love reading about your development process and seeing where these iterations take you. I can't wait to see what results from this new direction! Keep it going!!