Hacker News Reader

Comment by avaku | Hacker News Reader

Hi! I made this app in my spare time off work... I described the motivation on the website. Please let me know any feedback! Would really appreciate it.

View on HN

I'm curious how this differs from wavelet analysis? Is this a wavelet application with the wavelet function optimized by AI?

I figured out that I've implemented a real-time wavelet transform while I was doing this... :) I tried and couldn't find any good apps that would use it to display real-time analysis of sound, so I made one. Actually, I was working on the AI to generate music, but I haven't got to the AI bit yet, and decided to release this quickly, just for fun, see what people think. I think much better visualisations can be made from the output of this algo...

Could you describe more in detail how your algorithm differs from a wavelet transform, if at all? The website doesn't say.

You mention that your wavelet transform is "real-time", but the normal wavelet transform already is real time.. or what am I missing ?

I guess you're missing the detail of how it's actually implemented in practice. The short description would be just "wavelet transform" I guess. The details is in how to implement it efficiently so that it can run on the phone in real-time, and draw the visualisation at the same time, without drowning the CPU (plus the additional feature of high-precision detection, which is not solved by wavelet transform itself). I will describe it in more details when I'll be open sourcing the algorithm (after I've implemented something on top of it that I could potentially start selling).

I see, thanks. FWIW, if your phone has an APU or GPU, it is quite easy to implement with CUDA. Is pretty much a massively parallel problem.

That would be hard to target different phones... I am just using SSE/AVX (via Apple's Accelerate framework) for now for some parallelisation, helps to speed up the detection x10.

I like the bubble visualization a lot. Great work! Have you seen the art of mixing https://www.youtube.com/watch?v=TEjOdqZFvhY ? It seems like something like your tool would be really helpful for mixing music.

Haven't seen that video, thanks! I am a bit of a home DJ thought, in addition to my full time work in machine learning :) I've thought about DJ applications as well. It all comes back to the original goal of the project to do machine learning on music. One of the potential applications I've thought about is a new interface for mixing! I've imagined an app where you would "grab" different parts of songs/tracks and "drop" them into the mix, and remove some other parts. The ML algo would need to break down each track into its composable parts, and make them available for mixing with parts from different tracks. Possibilities are endless! :) That would require a lot of work (my dream job) :)

make it free and open source?

Ideally, I would like to make something else based on this algorithm that would allow me to leave my work first. Like maybe some fancy 2D/3D visualisations. I want to then open-source the core algorithm. Maybe I should do it sooner though. I don't know, any advice?

I will download your app later.

It’s interesting as I just worked on FFT and sound visualization with Python, Seems quite troublesome to circumvent the need for a sample buffer or time-windows.

If you see a way to monetize I would definitely do so. Only you can decide how unique your method is. If it is very unique, patent it? Perhaps Shazam or other apps could use your method?

It is really hard with FFT to get a good frequency representation. I've described a bit more on the app website why on front page. It's basically like deciding whether you want to detect the frequencies well, and loose time resolution, or if you want better time resolution, loose the ability to detect low frequencies :) What I've done is kind of like wavelet transform, but "real-time", and I have one extra feature for "high-resolution" frequency detection that goes in addition, and I've done the math for that myself, I haven't seen it anywhere... Wouldn't have time to do the patent at all, I have another main job :) Ideally, I would rather make something more impressive with it, and then "sell out" :)

Multi-Resolution FFT and Short-Time FFT (STFT) is commonly used to avoid the frequency/time resolution tradeoff. Example here https://ccrma.stanford.edu/~jos/sasp/Multiresolution_STFT.ht...

That looks good!

I would love to use your algorithm for a personal hobby project. So far I’ve ported other similar algorithms such as Colorchord but this seems like an interesting take on getting data out.

Being able write a lib to plug into TouchDesigner or OF would make it fun to write other kinds of visualisations.

I am definite considering releasing the algorithm, it's currently in C++. But I want to build something on top of it first, so that its application can attract more attention. I've tried releasing the "algorithms" before -- nobody cares if you first don't show something cool can be made with it...

I'm open to help write bindings or cool demos if you're interested.

thank you! I can clearly see the interest among people now! I will think what's the best strategy. Definitely be releasing the core algo open source, need to prepare, figure out what's the best way to do it. Thanks again!

Reply on news.ycombinator.com