Warning: potentially NSFW - language (for those still in offices)
The song in question is sung with a very clear voice, with minimal instrumental backing. That may suggest the limiting factor is Google's ability to parse songs, rather than users' ability to hum them.
Aha, so that's where I had heard it before!
'The song is best known under the alternative title "Sukiyaki". In Japan it refers to a Japanese hot-pot dish with cooked beef, the word sukiyaki does not appear in the song's lyrics, nor does it have any connection to them'
"An instrumental version of the song was played by NASA over the radio for the Gemini VII astronauts as mood music, thereby becoming one of the first pieces of music sent to humans in space."
Whistle and look up so you can't see these tears..
It managed to be a #1 hit in the US even!
For about two weeks I had this beautiful haunting melody stuck in my head. It would just turn on at certain times in the day and I'd find myself humming it. I have around 3500 CDs, mostly jazz but boatloads of classical stuff too. For days I rifled through bins of CDs (which is something I love doing) looking for the source. No dice.
Finally, a breakthrough: One day the song was playing in my head and I noticed the timbre of the instrument was a clarinet. And it was kind of raspy, which was weird, because the melody seemed more classical and classical clarinetists have super clean tone. I immediately knew to start looking for Don Byron albums, because he probably makes up 90% of my clarinet jazz albums. Sure enough, after a couple hours I found it. The last tune off the "Tuskeegee Experiments" album is a Schumann (!) piece called Auf einer Burg:
This isn't just "big tech replicating a commonplace feature", it's big tech implementing it at all where little tech could not.
This is a common fetishization on HN, but HNers forget that execution is all that really matters. Little tech demos do not. Though we on HN act like tech demos are the hardest part.
SoundHound is definitely not a tech demo.
Well I guess our experiences were quite different then. I was able to get SoundHound to recognize the vast majority of the tunes I hummed for it back in 2008/2009. And I used it a lot. I did only choose popular songs though. It's not surprising IMHO that a company 12 years later with much greater resources could extend recognition to a much larger variety of songs. So I wouldn't call it "little tech could not."
"Recorded in Berlin, June 17, 1941. This was based on an old Yiddish song 'Yossel, Yossel' (by Nellie Casman) which Sammy Cahn and Saul Chaplin later retitled 'Joseph, Joseph.' Wonder if the censors realized this."
So it was really not working at an acceptable level, or both me and my friends are terrible hummers.
Fun anecdote: I witnessed one user sing a song into SH so pitch perfect that (from the results) it had made a direct music match!
We often heard that Google were attempting to replicate this technology. They even hired key employees involved in this part of the app. Still we maintained that particular technical moat for 12 years!
Maybe I was humming something not included in the humming database then (I didn't know they were separate), or maybe, I'm just bad at humming :)
> It’s amazing how quick we are to go ooo ahhh when big tech replicates what is already common place in little tech or academia.
Translation: "You're dumb for liking this thing I had years ago". What an empty comment.
I don't have an Android and never heard of SoundHound, and I'm quick to go "ooo ahhh" because last weekend my mom was humming a song she didn't know the name or words of, and next time I talk to her I can tell her to try song humming on OK Google, and her phone now does this by default, and if it works I'm predicting she will tell some of her friends or sisters as well.
perhaps I should make sure to remind her that she shouldn't be impressed if it works, because a thing she would never have paid that much for, didn't care about and never owned, could do it years ago, would that help?
That’s just your extremely uncharitable and incorrect insinuation of my comment. What I was saying was not about the user’s reaction to the feature but instead about the admiration of the company. It’s not impressive if Ford uses a wheel on their car because a wheel is old technology that others have already done well.
Is it not impressive that Ford model T became the mass-market car recognisable a hundred years later, because “we’re too quick to ooh and ahh other companies have used wheels before”?
Look at the people in this comments thread saying it made their morning and they found songs from childhood and they’ve been experimenting with foreign songs, and I ask again, what does “I’m not impressed it’s been done before” add? Why do you think that’s a good, useful, comment? so what if it’s been done before?
If another company had come out with a low cost well-working automobile 12 years before Ford, then people should not be as impressed by Ford. That's my point.
What exactly is your problem with people being excited or impressed by something that was done before, which makes it worth a comment?
Again, you are completely misinterpreting my comment. Google is a trillion dollar company. It's not impressive that they recreated something a startup was doing 12 years ago.
> What exactly is your problem with people being excited or impressed by something that was done before, which makes it worth a comment?
For the third time, I'm not saying people shouldn't think the feature is cool. They should just IMHO not be heaping praise on Google for it since it's old tech, and perhaps the scrappy startup that did it first a long time ago deserves more of the credit.
Then I tried "Happy Birthday" and got three results:
"Las Mananitas" by Canticos
"Iyi Ki Dogdun Melissa", by Eser Ulun
"Iyi Ki Dogdun Zarife", by Eser Ulun
I'm no Pavarotti, but I can sing on pitch and have played the violin for 40 years.
EDIT: Heh, "Iyi Ki Dogdun" is Turkish for Happy Birthday, and the links would have taken me to YouTube videos of the song sung in Turkish, then English. But why not to videos in English? YouTube has some, and I'm an Anglophone user in the USA.
I'm not sure how many people think of hum the same way. I've noticed lately some words being used slightly differently than I am used to (the one coming to mind right now is "giggle" in a story that in context I would interpret as meaning "chuckle"), and at this point it's happening enough that I'm not sure if there's a shift in perception of these words or if I just always interpreted them more narrowly than they were intended.
I tried using my mouth open and didn't get significantly different results, though.
I'm pretty sure it's an instrument line, not vocal.
Also, why the hell do I need an app for this? Is it actually running a local model? Because my guess is that it's just hitting an API.
There's no way it would do much locally, but maybe they just wanted to make sure the audio passed to the API has a certain sample rate and encoding?
Most likely they are mapping the interval between the sung notes and using that as part of the ‘melodic fingerprint’ for matching.
For example, long press the home button, say whats this song, hum a song, it gives 3 suggestions, click a suggestion, click the YouTube link to hear it, click twice back to try another one of the three... And Google assistant has lost all state and no longer shows the 3 results. And if I "swipe up to see your snapshot" it crashes and dumps me back at the launcher...
Did they not even test this? Being able to do back and forward navigation without leaving the user on a random screen is like app design 101, and you can totally write unit tests for it.
As I write this from my iPhone, this might singlehandedly be the reason why I switch to a Pixel instead of another iPhone this fall. I was always disappointed that even after acquiring Shazam, Siri could never seem to correctly guess what song was playing unless I had a perfect recording.
Anytime I go to a live concert venue, or even a club, I never can get a recognizable song from my phone. Live music I could maybe sort of understand, but the club part failing was always weird to me, because besides a little bit of distortion, the song being played is coming directly out of a speaker. It shouldn’t be that hard right?
The button is there for me, and hum to search works.
When I first installed the app after reading the article, to test it out, I couldn’t find it even when I pressed the microphone symbol.
But after rereading the article I installed it again and when I went and tapped the microphone I saw the Search a song button.
Immediately after updating, there was no button. Maybe something changed on their end or something needed to sync in my phone.
edit: now the button does show up. Tried humming "Peter and the wolf" and it failed.
It's almost spookily accurate
Reminds me of Google Translate on photos/images, which has been available for years on mobile but never on web.
I hope this doesn't become a trend.
I mean I recognize that Translate on images in the browser - especially if combined with Youtube and Image Search - would be incredibly powerful, but it just wouldn't be as much a thing on desktop.
I'm pretty sure (don't quote me on this) that there's a bit of processing done on the device (character recognition / isolation) before sending it off to Translate as well; not sure how well that would port to a browser. Should work just fine with wasm though.
"song that goes meow meow meow meow meow meow", can't find it at the moment for a timestamp. Wonder how many real queries of that actually happen.
> I have the tune in my head of a metal song. but I can't figure out. It's goes like on guitar ha ha ha. ha ha ha haaaaaaaaaaaaaaaaa. ha ha ha ha haaaaaaaaaaaaaaaaaaaaaaaaa.
I was thinking it learns the tune of song and matches with that so language should not matter.
It worked for "Desert Rose" though.
Edit: Also worked for an Indian song.
Also worked for few other recent famous Pakistani songs. I guess the first song that I searched is probably not listened on YouTube as much as its known by heart.
It could not search for national anthem itself for the reason.
1) A "I'm finished humming" button
2) A way to tell Google what song I just hummed when it guessed wrong
It failed for me on "Sofia" and "Peter and the Wolf" -- maybe this only works for recently popular songs?
The Second Waltz, Johann Strauss.
Okay, a very famous piece of music, and a very recognizable melody, but I just thought my warble was so far off that most humans wouldn't be able to tell me what it was!
So, I bought SoundHound 8+ years ago in a Google Market promotion offering several apps for pennies (10c, I guess?). Tried it without much hope and damn it, it worked well. Sometimes it goofed, but truth be told, it may be my problem.
Also, it had the ability to identify covers in a crowded bar, or non-playback live versions. Shazam, which was very popular at the time, admittedly couldn't do any of that - in their FAQ (which I could not find right now) they even told that if Shazam recognize a live version, it must be either a playback or a player with microsecond precision repeatability.
Edit: After tapping the microphone icon, I have to tap the “search a song” button instead of saying “what’s this song”
* Everybody Has a Dream by Billy Joel
* Kids by MGMT
* Good Day Sunshine by The Beatles
I know I'm not the world's best singer, but I was humming the right notes.
It got MGMT's Kids no problem, and I know I sung out out of tune.
I'm very impressed.
I found out worked better if I use "doo dee doo dee doo" rather than humming.
It wouldn't surprise me if this leaned heavily on rhythm as a signal.
Tried with "The Most Mysterious Song" https://youtube.com/c/WhangWhangWhang/search?query=mysteriou... and "The Case of the Missing Hit" https://news.ycombinator.com/item?id=22700482
But then after I’d uninstalled the app I went back and reread the article and saw it say to press microphone and then button for search a song, reinstalled app and then I saw the button there.
Still don’t know if it was there or not the first time around. Maybe I just didn’t see it? Either way, it’s there now. Using iOS btw.
I was curious to see how it would fare because for example when I’ve tried to hum into Shazam or SoundHound in the past they’ve never come close to finding what I was looking for.
But I read in the article that Google is using machine learning and that they have actually trained their hum to search on recording of humming so I got my hopes up.
So in the Google search app I hummed a part of the melody for the song What is Love by Haddaway and... it recognized it! I’m impressed!
edit: works for me now. After a few minutes the button just started showing up
Also attempted Grieg's Hall of the Mountain King and got this, not quite what I expected but close enough: https://youtu.be/e1kwr-C78mI
It was an old web application where you tap the beat of the song you're looking for on your keyboard, and it presents you with a list of close matches. This was way back before smartphones and the AI revolution were a thing!
First I tried the Mario theme song, it got it, but with very low confidence and had about equal confidence for a Bone Thugs song lol.
Tried it again with a rock song that was at least popular enough to show up on an official Spotify Playlist for its decade and genre. It got that with pretty good accuracy.
Finally I tried some sludge metal and punk tunes. I can't vocalize the sludge sound worth shit so I'm not shocked it didn't come back with anything, but I'm mildly disappointed it didn't find anything for the relatively simple Choking Victim song I vocalized, but I assume song popularity is a big component.
Cool feature though, unfortunately I can't hold a tune for as long as the snippet it wants lol.
I think I see a future party game. No song repeats, first person to fail to get Google to come up with the correct song while using no real words loses.
Luttle luck with French songs, got weird matches that didn't sound at all like the original.
Zero luck with tunes from clqssical music, even with very well known arias.
May be an artifact of how this has been tested, and by whom...?
I am a non-native English speaker, and I have been trying songs from different countries. I did notice that English songs have a better match than songs from other regions and languages. I wonder if their training dataset has "overfitted" to such music, or is such music inherently represented by some underlying features that are better distinguished than others.
: E.g. English ("My Heart will Go On" and "Skyfall") fit with 78% and 85% respectively, while Japanese ("Tonari no Totoro") and Hindi ("Tum Hi Ho") fit with 42% and 48% respectively
You can also search by melody or by contour
What? Humming is not different across dialects.
I don't have time to finish it but i would love it if somebody would make this
I'd love for this to improve. I'm a jazz musician, and I know hundreds of tunes, but can't for the life of me remember their names when I'm on the spot. So it's hard for me to suggest a tune when I'm playing with a band.
It also requires to hum for relatively long, sometimes I remember like 5 seconds of a song but the app wants me to keep going for 15.
It also thinks the tinny reproduction of "Go to sleep little baby" my kids nightlight plays is "Get ur freak on" by Missy Elliott. Which I can definitely hear, but seeing it on my ambient display at 0300 one morning made.me question my sanity.
I'm wondering how did they construct the dataset to train ML models on it. Did they just recruit "labelers" who hummed for evey song? Or maybe do they have an automated way to detect humming in youtube videos, and later label those with the corresponding song?
Maybe this (no idea if its good) https://audiotag.info/
This is a very old joke that has the punchline of, "No but if you hum a few bars I can fake it"
I was rather whelmed when I saw it could just find songs that were playing right now.
I gotta update the app or install something on Chrome. Why not just make it available on the web somewhere?
I must have an older version of the app. Followed the instructions and it recommends Shazam!
As for those interested in trying it out, the feature is part of Google Assistant.