Music Search App
ginnyjollykidd .
I haven't come across any search engine that does this, and I'm surprised someone in YouTube hasn't made one available, yet. Maybe only us old farts with a music background want to be reminded of the titles and information we once had crammed into our heads that now ends up being only a litany of tune snippets that have no information attached anymore.
I want to create an app that will take a .wav or a file of a music phrase plunked out on a music- composing app and "Google" searches music files on the Web. YouTube and Vimeo would be great targets, and there might be an option to put key words in like genre, artist, or other keys.
Digital music or a digitized recording should have patterns for the passage that can be approximated in a way similar to Google Image search.
I realize this is probably a formidable project who has had only 2 classes in programming languages (i. e. me, and Pascal is useless and C nearly so), and I figure I have a lot to learn.
So I'm putting it out there to all you programming geeks:
What language is best to use for such an app?
What kinds of information would this app be needing in order to search audio/video online? How do I search for genre and artist keys or other key words? And how would I generate a list of issuing matches like a search engine does?
Thanks for your experiential advice.
Yours truly,
Ginny Jolly
Ben Hibben
--
You received this message because you are subscribed to the Google Groups "LVL1 - Louisville's Hackerspace" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lvl1+uns...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Jared McNeil
ginnyjollykidd .
That's a cool idea. Anybody got an idea how to program it?
ginnyjollykidd .
Thanks.
Greg Miller
the sample I gave it. So my guess is it does a simple speech
recognition and just does a text search for the lyrics. Doing an actual
match on the music would require some clever tricks as you couldn't rely
on an exact match since each file could have different losses due to
compression, and added noise. My guess would be the problem is too
difficult to practically solve using today's computing power if no one
has done it already. But if you want to try it, I'd think a good way
would be to pre-compute a Fourier transform over different lengths of
each song you plan to keep in the database and just keep the dominant
frequencies that come out of the Fourier transform. You could then use
that set of frequencies into one of the many machine learning classifier
algorithms to produce a set of candidate results
(http://rayli.net/blog/data/top-10-data-mining-algorithms-in-plain-english/).
If you've only taken two programming classes so far, you're probably a
really long way from getting anything working, but it's not a bad idea
to have such a project in mind as a driving force to learning, and it
might be practical by the time you are good enough to solve it.
As far as the language goes, the final result would likely need to
be in a low level language like C in order to produce results quickly.
But solutions involving such complex algorithms are rarely implemented
immediately in a low level language. Rather, people usually start with
something like Matlab or Octive to play around with different ideas, and
then ideas that appear promising are implemented in another language
like R or Python with SciPy and off the shelf machine learning APIs for
further testing. And only the most promising of those are implemented
in C, followed by a lot of bouncing back and forth between prototypes in
different languages to tweak and experiment with different things until
you find something that works.
Even if you never produce anything that works, you'll have learned
a lot along the way that can be applied to other problems.
On 8/17/2015 3:23 AM, ginnyjollykidd . wrote:
>
> That's a cool idea. Anybody got an idea how to program it?
>
ginnyjollykidd .
I knew it would be formidable. At least I DO have C under my belt (and a book or two to guide me). Looks like I'll have to learn quite a bit about digitizing music. Any music geeks out there?
Ginny :-)
bruce
to "search" for similar music (or the same) based on the initial
sound/music file possessed by the user.
Ie the user would upload a music file, and the app/algorithm would
then more or less spit out data about the song, or find similar songs
to the input song,. Is this correct so far? Is there more/less to the
original OP's question.
If the goal is to be able to get "similar" music, or any meta-data
regarding the initial music file, if you have a couple of mill it can
be done! This is somewhat similar to what companies like
Pandora/Spotify/etc do when trying to figure out the playlist to
generate for a given user.. Their approaches are way more
sophisticated, but it's on the same track.
The issues you run into are similar to issues one would run into if
you were going to develop an app to find" pictures similar to an
initial picture.. The attributes that one could use to target as
important criteria are almost infinite. how you reduce that group of
attributes, and then how you build a classification process is where
the magic occurs.
Ben Hibben
Patrick McCarthy
This actually exists. (Not sure how you get your hands on it though)
It’s all nicely patented by google & company.
Use:
Company A grumbles that You Tube has peeps scamming Our Musical Stuff…. And they enclose copy of said noise.
YouTube creates a kind of musical thumb print & actually goes out and looks at all the butzillion files it has, looking for a match
If(when) it finds a match, it sends notices to Company B, C, and D that they are infringing
For PEOPLE… little old people…. It just blocks the entire file & replaces it with a “This dud(ette) is a jerk & used copyrighted material” placard
This actually works VERY well. Not just on the actual music, but on variations. losses due to compression, and added noise do not slow it down. (In some cases, it works too well… there have been cases of the placard going up cuz some dude in the background was HUMMING a tune…. And it got caught.)
From: lv...@googlegroups.com [mailto:lv...@googlegroups.com]
On Behalf Of ginnyjollykidd .
Sent: Monday, August 17, 2015 12:06 PM
To: lv...@googlegroups.com
Subject: Re: {LVL1} Re: Music Search App
I knew it would be formidable. At least I DO have C under my belt (and a book or two to guide me). Looks like I'll have to learn quite a bit about digitizing music. Any music geeks out there?
Ginny :-)
……………………………………………………………………………………………………………………………………………………………………
Patrick McCarthy,
MIS Software QA
Samtec New Albany
Office:
812-944-6733 X:2713 | Mobile:
(502)-939-1756 | Fax:
(812)-981-4359
www.samtec.com
ginnyjollykidd .
So it exists in some form. Is the patent available to look at? Surely it can be reverse engineered legally, right? The look and feel would be different, I think.
Ginny :-)
Patrick McCarthy
From an oldish blog post:
YouTube narrowly avoided legal trouble by promising the big media companies that they’d develop a system that could detect and automatically remove any copyrighted material that was uploaded to the site. But in reality, they didn’t actually develop the audio fingerprinting system; they licensed it from a company called Audible Magic.
From a deep dive into GitHub, I found the following… grab a fork and tuck in J
https://github.com/echonest/ All of the items in this nest are based on this: http://echoprint.me/
Finally, if all you really want is a GOOD app…. Here’s a largish collection (10)
From: lv...@googlegroups.com [mailto:lv...@googlegroups.com]
On Behalf Of ginnyjollykidd .
Sent: Monday, August 17, 2015 1:54 PM
To: lv...@googlegroups.com
Subject: RE: {LVL1} Re: Music Search App
So it exists in some form. Is the patent available to look at? Surely it can be reverse engineered legally, right? The look and feel would be different, I think.
Ginny :-)
……………………………………………………………………………………………………………………………………………………………………
ginnyjollykidd .
Thanks!
Ginny :-)
galenpowers
It was able to get Quintette avec piano, Op. 14:: III. Presto while listening to the tv in the living room.
ginnyjollykidd .
Well, cool!
Not exactly what you suggest, but fun, free and easy. On an Android device with google voice search enabled, say "ok Google" "what is this song"
It was able to get Quintette avec piano, Op. 14:: III. Presto while listening to the tv in the living room.
Greg Miller
they comprise a whole other field of machine learning. Yes, they're
complicated, but there's still quite a bit of information out there
about them, and probably even open source libs implementing different
ones. There's actually a free Coursera "at your own pace" course on
them: https://www.coursera.org/learn/recommender-systems . I've watched
some of the lectures and they seem pretty good, one of the examples
worked throughout the class is exactly predicting what music a user
would like base on what they and others have liked. I don't think
that's what Ginny was actually asking about, but might be of interest as
it's probably a lot more approachable of a problem.
ginnyjollykidd .
Yay! That's great! I think I'll have to plug into that!
Ginny
Greg Miller
has a couple of PDFs at the bottom that would be good pointers in the
right direction:
https://en.wikipedia.org/wiki/Acoustic_fingerprint
ginnyjollykidd .
Cool!
Back to the original topic, the Wikipedia page for audio fingerprinting has a couple of PDFs at the bottom that would be good pointers in the right direction:
https://en.wikipedia.org/wiki/Acoustic_fingerprint