Show HN: Galiboo – An A.I.-powered API for music

[subby]

Hi guys! At our startup, Galiboo, we've created an AI-powered API for music, which can do things like:

- searching for music with natural language - extracting hundreds of moods & emotions from any track - finding tracks by tags - get similar-sounding music - and more.

The website is https://galiboo.com/, and the API docs are at apidocs.galiboo.com. We've also got a live demo at demo.galiboo.com.

We also have a Python client library at https://github.com/galiboo/galiboo-python.

Our APIs are in beta, so we'd love to hear your thoughts & feedback! :)

Thanks!

[NegatioN]

Cool API, I'm probably gonna try playing a bit with this when I get time :)

How big is the selection of songs you're training on and can recommend from or search through (non-demo version)?

How diverse are the languages? Is it mainly English since it offers a text-search?

To have a non-demo API-key, how much do I have to commit? Is there a "pay-as-you-go" model where I can use this as a personal thing, or is the only target audience companies?

Edit: Found the answer to the first question for demo-users in docs "For users with demo API keys, we've loaded a fairly diverse, yet small, catalog of about 20K+ tracks into our backend system"

[subby → NegatioN]

Hey, thanks for the feedback!

We're working on a pay-as-you-go model for developers & personal use, which we'll have live in the coming days.

Until then, you can use a demo API key that we're currently in the progress of emailing out to everyone that had requested one. Thanks! :)

[subby]

Hi guys! Sorry for this delay!

Due to the immense feedback & unexpected surge in API key requests that we received from being featured on Hacker News, we've decided to open up access to our API platform to all developers! :D

So now, you can get your own API key at: https://galiboo.com

Our APIs are currently in early beta, so we'd love to hear any feedback that you might have! :)

Our API docs: apidocs.galiboo.com Our Python library: https://github.com/galiboo/galiboo-python

You can also join our LIVE chat (with us & other developers) at https://gitter.im/galiboo/Lobby

If you have any questions, please shoot me an email to subby@galiboo.com and I'd love to help! :D

[peterlk]

How are you guys better than/different from Echonest? Can I give you new songs and get back a tagged blob of attributes?

[subby → peterlk]

Hey, thanks for the question!

The main difference & significant improvement that our technology provides is its ability to not only extract high-level data (e.g. emotions, tags) from music (Echonest, for example, primarily emphasizes low-level attributes like beats, tempo, etc.), but also perform other important operations over this data (like querying by moods/tags, finding similar tracks, search by natural language, etc.).

So, in a way, our technology acts as an actionable intelligence layer over music, as opposed to just extracting & returning low-level data.

And yes, you could give it new songs & get back the extracted metadata. Plus, our commercial users can also integrate their own music catalog with our technology.

Thanks for the question! :)

[MediumD]

This is really cool. I've wanted something like this for a while and have flirted with the idea of building something like this myself a few times. Great job on shipping something! Can't wait to play around with it.

[subby → MediumD]

Hey thanks a lot for the feedback! :) The demo service seems to not be working temporarily, but we're on the issue and we'll fix it fast. Thanks again! :)

[subby]

Hi guys! Seems like we've received a load of requests that we weren't expecting, so our API servers seem to be suffering temporarily. We're working on it though! :)

[davedx]

Nice open source stuff too.

https://github.com/galiboo/olympus

[subby → davedx]

Hey thanks for the feedback! :)

[onion2k]

I tried it on my favourite demo scene video[1] from the last few years, and it did really well at tagging it. That's cool.

[1] https://demo.galiboo.com/analyze/?url=https%3A%2F%2Fwww.yout...

[subby → onion2k]

Thanks for trying the demo and giving us feedback! :)

[afro88]

Looks cool, but I noticed that a lot of videos have both "major key tonality" and "minor key tonality" in the tags. Maybe you guys need to have a higher confidence score for a tag to be real / less tags? Seems to be a bit of a mixed bag at the moment.

Dying to try it on some lesser known songs when the traffic dies down :)

[subby → afro88]

Hey, thanks for the feedback! We're currently experimenting with a few different techniques for this. We're actually using a lot of tags because the internal vector representations that our A.I. creates for these tracks become very accurate with this large number of tags (which helps with finding similar tracks, etc.). Thanks a lot for the feedback! :)

[subby]

Hi guys! We think we've fixed our API outage issue, so our APIs and demo should be available now, and be able to scale with demand. Hopefully everything goes well! Thanks for the patience & feedback guys! :)

[shusson → subby]

the demo is still not working for me...

update: it's working

[jrjames83]

Very nice - can't wait to try. What kind of features and distance measurements do you use for the audio similarity? I've worked on that quite a bit in the past.

[hemulin]

That looks really awesome! Do you have any estimations for pricing plans? It's mentioned that the demo contains ~20K tracks, how big is your "live" data set?

[subby → hemulin]

Thanks for sharing your thoughts! We're currently creating customized pricing for each of our enterprise customers, to better suit their needs. Our live, commercial catalog consists of a growing catalog of about 64M tracks. :)

[faitswulff]

A naive question, but how hard would it be to generate original music from a natural language search instead of just returning matching tracks?

[subby → faitswulff]

Oh, that would be really hard for a computer to do, but we're thinking of trying that after we nail this one. ;)

[verrecken]

"Learn how the music in your ads affect conversion rates, find the right music for ads, and more." I dont get this. How does it work?

[subby → verrecken]

Hey, thanks for commenting! The music typically used in ads (like video ads) can have a great impact on click-through-rates and conversion. So, by testing ads with different tracks by specific moods, we can find the right music that optimizes ads. We're currently testing this with a few private beta customers, so we'll see how it goes! :)

[adamhepner → subby]

I can almost imagine a service that allows you to do multiple versions of your video ad, with script changes, various music, graphic styles, etc. and verify the rates.

If somebody ever runs with this idea - just let me know afterwards how much have I missed on?

[devxpy]

How does this compare to Spotify's recommendation and audio features API?

[sobradovic]

Great idea. One thing that I have noticed, demo does not work. Take a look :)

[subby → sobradovic]

Thanks for the note! We're currently scaling our server capacity, so it should be resolved soon. Thanks again and sorry for this! :)

[pags]

Works well. How much are you pulling from the video title and description?

[subby → pags]

We only use the video's audio, not any other attributes (like text, title, description, & even the video itself). Thanks for the feedback!

[bfuller]

ah this is great thank you.