We're working on something that will change the way you use your sample library.

[timnoric]

Yeah that would be cool! We're hoping that Atlas will be a first step in that direction, and say if you have a bassline that is really heavy, you can use Atlas to easily find and replace an overbearing kick with one that is bright and snappy etc. ie bright kicks are top of the kick cluster, boomy ones at the bottom.

It's an interesting question though and we think about it a lot here at Algonaut. It is possible , but how much would you want an AI to make the music for you? Is it still 'your' music?

[raintalk]

The concepts, ideas, and choices are still mine. Just make a "recommendation".

I want it at this product level. For Mixing and mastering there are products like Izotope N2O8, smartEQ, and side-chaining methods ... but this is all too much work at the end.

It'd be better for me if the AI is picking better samples, or at least suggesting it as I'm working through things. Like "80% would pick this sample of this one in this case because ..."
Maybe I choose your advice because I'm not trying to make a personal statement in that section. Maybe I ignore your advice because I don't agree. Allow for an exception in this case. (don't nag me again)

You have all the information from all the samples. So Suggest something when I ask.
I'll ask at certain times. I'll import a bunch of samples that I haven't even listened to . You analyze them and make some suggestions.

[Rajiv]

Hi - I really like this, particularly if the location of the dot on the grid, and/or the color, size, and shape of the dot communicates useful information about the sound. Like maybe dots lower in the matrix have a lower pitch, maybe the color reflects electronic vs. acoustic timbres, that sort of thing. The user could customize how properties of the sample can be reflected by the available visual metaphors.

How about expanding this for use with loops, as well as samples? I keep almost buying RyAudio's Sample Librarian. It's good, but it's still a big long list of files, and it doesn't have (or I didn't find) any ability to auto-tag files. A solution that identifies properties of loops (key, bpm, rhythmic vs. tonal, busy vs. sparse, high pitch vs. low pitch), pulls in tags from file/folder names (e.g., instrument, sample pack), allows me to apply my own tags, AND THEN creates a visual matrix for me to audition loops in the context of a track I am working on... well... that would be simply spectacular.

Good luck... I think you are really onto something.

Steve

[deastman]

I think this looks very promising! Great idea...

[dandezebra]

Looks really cool! I have so many questions - some that others have asked and some not! I'll wait, however, and see what you guys start sharing.

[timnoric]

Looks really cool! I have so many questions - some that others have asked and some not! I'll wait, however, and see what you guys start sharing.

Im keen to hear your questions now! This is why we're sharing early dev snippets!

[trusampler]

I love the idea, I just hate the UI, seems like I'd be mousing all over the place to find samples, when I already have them in certain folders, seems like it'd be cleaner with a layout like Akai's VIP, but thats just my opinion, good luck with it,
I really hope other developers can start making these sort of apps happen and always in the form of vsti/au as the daw makers have simply lost it mostly when it comes to browsers besides maybe steinberg with mediabay. Presonus where the hell are you?

[VariKusBrainZ]

Potentially game changing

It could analyze a huge pool of samples and you could filter only those relevant to your needs, with no human analasys or interaction.

And what about analasys of samples similar to each other so you could intelligently interpolate from one genre of samples to another, mind blowing if you have a huge samle pool

[bob bobwood]

Personally I'd like a bit of playability, perhaps along the lines of Soundtorch;
https://web.archive.org/web/20140316181 ... torch.com/

[cron]

Wow, this is very interesting stuff. Will we have access to individual 'characteristics' when sorting? I'm guessing this is using some sort of music retrieval algorithms, so it'd be interesting to isolate one/several 'characteristics' (e.g. individual MPEG 7 timbral descriptors) in addition to 'overall' similarity. Lots of interesting implications for organising material in electroacoustic music there.

Keeping an eye on this!

[audiosabre]

I really like the idea, and the look. Any chance you could demonstrate it with a few hundred thousand samples, instead of a few hundred? My one shot directory contains 480,000+ samples

Also, how would one go about finding similar sounds, if the sample wasn't originally loaded within Atlas? I'd expect there to be a search function to locate a specific sound (and clearly highlight it on Atlas' map), but I don't see it in your features blurb.

Thanks, and good luck with this project.

[dandezebra]

Most questions that I have are bring asked.

Can you control the parameters used to sort? As in on the XY axis? What parameters will be available?

[Gamma-UT]

Also, how would one go about finding similar sounds, if the sample wasn't originally loaded within Atlas? I'd expect there to be a search function to locate a specific sound (and clearly highlight it on Atlas' map), but I don't see it in your features blurb.

Looking at the UI, I'm assuming it's using some sort of clustering to analyse the sounds so I imagine the software has to index all the samples it's expected to find.

IIRC there was some work on 'fingerprinting' techniques in academia (Queen Mary College?) that you could attach to music samples to speed up that kind of process as kind of audio 'tag' to avoid having to process each sample on the first pass. But you would have to load the sample in a system like that: it's just a database with a bunch of pointers to the actual samples.

This is without having any idea what Algonaut is actually doing, so this might be off target. But that map made me think "clustering" immediately. It should work reasonably well for one-shots. Loops and multisamples would be more...interesting to work with.

[audiosabre]

I desire the ability to pinpoint a dot in 'the cluster' (light it up baby!), then proceed to work directionally to audition similar or contrasting sounds. I'm also interested to hear more how the directionality now works or could possibly be configured. I liked what I read concerning the X/Y thing; the ability to input characteristics or attributes, then use directionality as a contrast/blend sort of tool.

Is there reason to how the titled cluster groups (the text displaying 'kick', 'snare' etc in teaser vid) are originally arranged/mapped, or could they potentially be user-dragged around to change the spread pattern of the individual dots/pixels from cluster-to-cluster? I hope that question made sense, lol. Could the map/arrangements potentially be saved, randomized, etc?

Adding to what another poster said about possible sample recommendation; it could conceivably learn about your taste through the samples you choose to drag out of the app. Scary and thrilling.

Do love this concept. Standard tagging systems are a bit hit and miss for me, depending on the product/dev. Alternative approaches to the sample selection process are welcome, and a truly exciting prospect.

Right. I ain't writing no novel. Have a musical week

[deastman]

This works well for one shot drum samples, but how do you handle longer recordings of a minute or thirty minutes or several hours? If you’re scouring my sample library, there will be plenty of unedited longer recordings and field recording sessions mixed in.