Google Hum to Search review: Finding 20 songs in 20 languages

1
Google Hum to Search '20 language' review: Shazam what?!


Let’s set the scene:


You know Shazam? You know how it can’t recognize a song if you were to sing or hum it? Well, Hum to Search was born to fix that. Apparently, it can identify a song by listening to the melody, your humming, singing, or even if you whistle. Bold claims! We had to find out whether they check out.

Preface


First of all… That's hands down the most fun day in the office for me (I'm actually working from home). I began this adventure thinking I'll have to hum away a few songs, just so I can confirm Google's Hum to Search is… OK. But! Was I surprised and drawn into it…

I ended up humming, whistling, mumbling, singing, beatboxing, making a guitar impression, playing the harmonica, the ukulele, and everything in between.

Oh, and I did it in 20 languages  because Google says the feature is available in 20 languages (although, does humming come in different languages?). Now, the tech giant doesn't doesn't say which languages, so I had to take the matter in my own hands (or mouth... ew!).

So, I figured I'd just go for the 20 most spoken languages worldwide. I've excluded two of the languages spoken in certain parts of India, so I can get a more diverse group of languages in. Of course, Hindi is still there.

So, let's see how it all works. If you don’t want to know about the "magic" behind Hum to Search, you can scroll down to the fun part - “The 20 languages experiment”.

How does Hum to Search work





When you hum a melody into Search, Google’s machine learning models transform the audio into a number-based sequence representing the song’s melody. The models are trained to identify songs based on a variety of sources, including humans singing, whistling or humming, as well as studiorecordings.

The algorithm also accounts for all the other details, like accompanying instruments and the voice's timbre and tone. What we’re left with is the song’s number-based sequence, or the fingerprint.

Then, the sequences are compared to thousands of songs from around the world and identify potential matches in real time.

For example, if you listen to The Weeknd’s Blinding Lights, you’ll notice that the song is recognizable regardless if it’s sung, whistled, or hummed. That makes it easier to find.

Hum to Search builds on Now Playing - a music recognition software released with the Pixel 2 in 2017, using deep neural networks to bring low-power recognition of music to mobile devices. In 2018, this tech made it to SoundSearch as a feature in the Google App and expanded the reach to a catalog of millions of songs.

According to Google, Hum to Search takes it to the next level because “it doesn’t need lyrics or original songs… but just a hum.” Or does it? Let’s find out!

The “20 languages” experiment


English

English songs made the experiment more diverse because I know the lyrics to some of the songs I tested, which unlocked my creativity! I figured I should not only hum or whistle but also try to beatbox, play an instrument, etc.:

  • Blinding Lights (The Weeknd) - this was too easy. I didn't have to sing. Google got the lead melody.

  • For Elise (Saint Motel) - not as popular as the first one. I figured I'd sing a little bit. Success.

  • Humble (Kendrick Lamar) - I'm very much into the keyboard that creates the rhythm in the song, so I tried emulating it with my… voice. I did my best, but Google couldn't figure it out. Some "rapping" helped.

  • I Spy (Mikhael Paskalev) - I tried playing the harmonica part of one of my favorite songs by Mikhael Paskalev - "I spy" on a… harmonica. It wasn't a huge surprise that Google didn't get it. However, it was more surprising when I played a recording of the artist himself playing the harmonica, and Google failed to get it again. I gave it one last try. I played (my) recorded harmonica part, sang the words over it, and... bingo! Takk, and Blagodarya, Google! (Mikhael is half Norwegian, half Bulgarian).

  • Drop It Like It's Hot (Snoop Dogg) - I couldn't skip this iconic Pharrell beat. Beat, because I tried beatboxing with all the mouth popping sounds and all that goes into it… Nothing.

  • Eastside by benny blanco, Halsey & Khalid - I chose this one because it has a very simple, popular, and recognizable string sound. Yes, I played the ukulele - no singing, humming, or anything else. Sure enough, it was a piece of cake for Google! I have to mention that I don't know the chords, and I'm rather bad at playing the ukulele, but the software excelled despite that. Impressive, given that I didn't help it by making any other sound (or maybe that helped?!)

  • For the last one, I was curious to see if Google will be able to recognize a song and then link it to the podcast it comes from (as a theme song). The song was We're Going to Be Friends by The White Stripes, which is a theme song on Conan O'Brien Needs a Friend. The song was recognized (singing), but Google didn't come up with the podcast. Perhaps I was asking too much, but this might be a great addition in future versions of Hum To Search.


Mandarin

So… At first I tried outsourcing the job to a friend who's from China. She said she was embarrassed to send me a recording of her humming - later it turned out she was still willing to do it, but at this stage it was too late. I tried mumbling it out, but it didn't work. I had to sneak a few of the Mandarin words in, and voila! I was able to find "Tian Mi Mi" by Teresa Teng. Xiè xiè,Google!


Hindi

This one was very easy. I already knew which song I'll try out, as it was stuck in my head for some time. I'd found a snippet of it (remixed) by chance while fishing for sound design samples, probably about half a year ago. The original song is "Jeena Jeena" by Asif Aslam. Google passed the test with flying colors. Jee shukriya, Google!

Spanish

Nowadays, Spanish is almost as popular as English when it comes to music. I went for “La Bicicleta” by Carlos Vives. Google came up with the original and Shakira's version. Muy bien! "Tu Me Dejaste De Querer" by the Spanish singer C. Tangana also wasn't a problem. Gracias, Google!

French

I knew exactly what to do: make the saxophone and then bassline sounds from Stromae's "Alors on danse", but this took me nowhere. I had to hum and mumble, so I went for “Tous les mêmes” and boom! 70% match. Merci, Google!

Arabic

Under different circumstances, I would have been challenged here, but I knew just the song I wanted to find because I'd overheard it many times but never knew its title!

So, this was a real one - I was actually curious to know the song. I started humming and mumbling the words of "Boshret Kheir" by Hussain Al Jassmi. The song is very popular in parts of Europe, thanks to the Arab-speaking immigrant communities. Now, I finally knew its name! Shukran, Google!

Bengali

At this point of the experiment, I was openly questioning my life choices. The most challenging part was finding a popular Bengali song. I'm not saying there aren't any. It's just that I don't listen to much Bengali music (I do now).

The song of choice was "Tomake Chai" by Kabir Suman (beautiful voice!). I tried humming, but it didn't work. I had to learn the hook… Still - no luck. I played the actual song, and Google found it immediately. So, either I'm not doing a great job with Bengali, or the language isn't amongst the Hum to Search languages. Regardless, dhonnyobaad, Google!

Russian

I was tempted to do Gorky Park's "Bang", but then I realized it's in English (duh!). So, the song was "Katyusha" by Varvara - very popular. Over 50% match on the second try (humming). Spasiba (Спасибо), Google!

Portuguese

I actually have a pretty basic understanding of Portuguese, and I can probably get by in Portugal for a few days. So, that gives me a slight advantage. First, I went for Salvador Sobral's "Amar Pelos Dois". 50% match on the first try. Jobim's "Aguas De Marco" (whistling) was too challenging for Google, even though I picked a distinct part of the song where the melody is very recognizable.

I moved on to Tego Calderon's "Pa' Que Retozen", as the lead synth is extremely recognizable. No luck with whistling or humming. I even tried helping it by beatboxing along (with the melody on top!), and it still came up with nada.

I knew I had to go popular with Don Omar - "Danza Kuduro". Again, no success when trying to hum or mumble to the opening pad. Finally, singing helped and I scored a 40% match! Obridago, Google!

Indonesian

I was very intimidated before going into this one. It turns out my Indonesian (humming) is better than my Portuguese. Success on the first try! I felt like I'd won Eurovision. Song: "Putus Atau Terus" by Tami Aulia. Terima kasih, Google!

Urdu

At this stage, I would have felt sorry for my neighbor if he wasn't a pain in the…

Urdu is spoken in Pakistan. The song I picked was "Pakistan Gayi" by Aima Baig. I really didn't know where or how to start here. After I stopped laughing at myself (and dancing, as the video was quite encouraging), I managed to learn to hum the catchy chorus. It took 3-4 attempts, and I couldn't believe it worked, but it did. Shukriya, Google!

German

I'm throwing a curveball at Google here. I'm in Germany at the time of writing this, and I'll go for a song, which was extremely popular in the country of sausages and beer about 6-7 years ago. I don't think this song was ever officially released. It's just 1:30 minutes-long. Still, it's on Spotify.

I had to be persistent, but it paid off! I found Kazim Akboga's "Is Mir Egal" on the second humming. He was a German comedian and lyricist of Turkish descent. The song went viral back in 2014 due to its funny lyrics and video. The more you know! And... Danke, Google!

Japanese

I made it a goal of mine: I'll sing in Japanese (no humming). If we look past the fact that it took me five minutes to find the lyrics of the song, it went much better than expected. Let me tell you... If I can mumble out a song in Japanese, anyone can. What I wasn't expecting is success on the first try! The song was a modern pop track called "Kirari" by Fuji Kaze. Arigato, Google!

Swahili

It was about time we made it to a new continent! Swahili's one of the official languages in East Africa (Kenya, Uganda, Tanzania, Rwanda, Burundi, and South Sudan). Here's the moment to say "asante" - "thank you" (in Swahili) to Wikipedia too!

I'm back to humming/mumbling because, let's be honest: If you're a tourist in any of the countries on the list, how likely are you to work out the language, let alone to start singing the catchy song you heard in the cab on the way from the airport…

The song was "Valu valu" by Jose Chameleone. Humming took me nowhere. I was challenged. Google was too. It guessed Daft Punk, Nicki Minaj, and even Florida Georgia Line. I tried whistling and even singing (if you could call it that) - nothing. Samahani! (Sorry!)

Tamil

I sang in Tamil, although my rendition was closer to gibberish. Shocker! First time's a charm. My research showed that "Kutty Pattas" by Ashwin was the number one song on "India's most authentic Weekly Music Countdown" - Radio Mirchi. Nanri, Google!

Turkish

You've gotten your vaccine (or negative COVID test); you are off to Turkey for a week. It's a night out - you hear "Marlon Brando" - a song by Zaynep Bastik, and you can't stop humming the melody. What do you do? You fire up Google!

I chose this song from Spotify's Top 50 because it's very melodic, and it's easy to "oh-na-na-na" it. In the end, it took about five tries of humming and mumbling. I had to give up on the chorus and go back to the first verse, but it paid off. Teşekkürler, Google!

Korean

With a risk to piss off the BTS Army, I went for "Lilac" by IU. I thought it was melodic and perfect to hum or whistle. I was wrong. I switched to "Celebrity" by the same artist, and after multiple attempts, I managed to get it right by making a poor impression of her voice in falsetto. Gamsa, Google!

Vietnamese

I tested proper humming (all the way). The song was "The Playah" by Soobin Hoang Son & SlimV. It has a very Uptown Funk-like opening harmony, which is probably why Google took it for Bruno Mars' song, and that's the closest it ever came. Despite that, Cảm ơn, Google! (yes, "thank you" in Vietnamese sounds like "come on" which ironically is the more appropriate expression to use after Google failed!)

Italian

This one was very easy for me, and respectively very easy for Google too. If so far I've had to prepare and learn the melodies of some songs, here I knew the song very well. I "sang" the chorus of Mahmood's Rapide, and Google got it right away (30% match). Grazie, Google!

Thai

Last time's a charm too - 69% (wink,wink) match on the first try! Song: "Just Being Friendly" by Tilly Birds & MILLI. This one must've been easy because the singing is very recognizable. Khaawp khun, Google!


Bonus (NOT!)

Korki Buchek (Корки Бучек) - "Bing-bang" (from "Borat: Cultural Learnings of America for Make Benefit Glorious Nation of Kazakhstan)". Unfortunately, not a great success. Still... Rahmet, Google!


Verdict


It's important to note that sometimes you simply need to channel your inner pop star and let your voice get melodic (even if it doesn't sound ideal). The melody is key, especially when you don't know the words, but in some cases it's actually the opposite, and if you don't know any of the words, you won't get to your final destination.

Hum to Search is not the only, but it's probably the best “hum-enabled” alternative to Shazam, as it does what Shazam does, but also so much more! SoundHound has pioneered this brilliant function, but Google's is more accessible , and it's likely more powerful thanks to Google's massive data sources.

Still, it's nowhere near perfect. As suggested throughout the review, we'd love to see it improve with rhythm, beat, and separate instruments. Linking a song to its respective podcast or programme would also be a great bonus! Also, we'd love to see the algorithm get better with melodies alone, without the help of actual singing/lyrics.

In the end, I'll have to give it to Google - this is a brilliant piece of software, and it shows how powerful, but yet intuitive to human needs tech is getting. I challenge Shazam and Apple to join the race!
FCC OKs Cingular\'s purchase of AT&T Wireless