Nice and very responsive, well done. I searched for "Imeprial March" (the John Williams Wiener Philharmoniker version) and this was one of the highest ranked results: Inauguración del Ferrocarril (by Nathan Stornetta). I kind of chuckle thinking about what a happy empire this would be with that theme song. And of course it also answers my initial research question...which national anthem is highest ranked for Imperial March. The answer is Taiwan...make of this what you will. Star Spangled Banner is the second one a bit down the results list.
Show HN: I trained an AI model on 120M+ songs from iTunes
301–310 of 444 posts
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#302search is not working for me
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#303https://maroofy.com/songs/724345280
That's just the same song always 6 times in a row.
you probably already know, one would prefer to have removed versions of the same song by the same artist at the very least.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#304It's quite good. Tried a couple house/electro/ebm tracks and it gave a good mix of stuff you'd expect and novel stuff you wouldn't. It finds a lot of songs that sample a given song, for obvious reasons. Glad someone gave this a go without any collaborative filtering or user ratings, just digging through waveforms alone.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#305It's a great idea and I like it a lot. Especially the execution. It loads nearly instantly and has everything I want there.
But of course I tried it with 2 very difficult songs I never could find anything like it: Desert Rose by Sting and Zoolook by Jarre. For Desert Rose it found some similarish things (but of course failed to capture the essence, what makes Desert Rose so unique) but for Zoolook it completely fell apart.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#306Search doesn't work in a recent Firefox on Windows - typing anything in the search box shows the "Loading..." dropout below it and then nothing happens. This generates no network activity and the console shows:
Cross-Origin Request Blocked: The Same Origin Policy disallows reading the remote
resource at https://cdn.segment.com/v1/projects/F4GFNelOpRsgUJc6iwTuiXr2t6AH5LCY/settings.
(Reason: CORS request did not succeed). Status code: (null).
NetworkError when attempting to fetch resource -- _app-f73fb5ecceb5fc1e.js:1:82295
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#307It doesn’t appear to work in Firefox. Looks like CORS is blocking a call to segment.
Same issue in Firefox
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#308Search doesn't work in a recent Firefox on Windows - typing anything in the search box shows the "Loading..." dropout below it and then nothing happens. This generates no network activity and the console shows: Cross-Origin Request Blocked: The Same Origin Policy disallows reading the remote resource at https://cdn.segment.com/v1/projects/F4GFNelOpRsgUJc6iwTuiXr2t6AH5LCY/settings. (Reason: CORS request did not succee…
Same goes for Vivaldi... I thought it might have received the famous "hn-hug-of-death".
//Edit: Just read the edit regarding ublock origin. So this website definitely needs a Privacy Policy then.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#309Lot me try this with Lotuk by Arsenal. Every other service will only recommend other songs from Belgium, disregarding any other form of similarity (to my dispair).
Currently, the website seems down though?
EDIT: the results are ... not what I expected. They are similar in some sense, but I wouldn't consider them "musically" similar. It's like each of the recommendations has something similar, whether the drums, the baseline or the voice. But none of them feel similar in the whole.
It's as if similarity is measured with an L2-norm in a high-dimensional embedding space, instead of cosine similarity? Did you experiment with different scoring metrics in semantic search? I would recommend cosine similarity, if this is an embedding space trained with neural networks, whether with a contrastive loss or with a gaussian prior.
Re: Show HN: I trained an AI model on 120M+ songs from iTunes
#310One of the files doesn't load right now
https://cdn.segment.com/v1/projects/F4GFNelOpRsgUJc6iwTuiXr2...
Error code: SSL_ERROR_RX_RECORD_TOO_LONG