Why Google Home Has Hard Time Recognizing The Smash Hit 'Despacito'
Why Google Home Has Hard Time Recognizing The Smash Hit 'Despacito'
The Latin Grammy nominated song "Despacito" is a smash hit on YouTube. But if you ask Google's personal assistant to play "Despacito" it can't understand the command. Even though YouTube is part of Google. It's a learning moment about artificial intelligence, how natural language processing works and why machines still struggle with translation and foreign accents.
KELLY MCEVERS, HOST:
Tonight at the Latin Grammys, the hit song "Despacito" is up for four awards, including record of the year. Clearly a lot of people know this song. But it turns out the Google Home personal assistant does not. NPR's Aarti Shahani explains.
AARTI SHAHANI, BYLINE: This has been the maddening conversation in my kitchen.
OK, Google, play "Despacito."
COMPUTER-GENERATED VOICE: My apologies. I don't understand.
SHAHANI: All right. Maybe that was too ethnic. I go gringo.
OK, Google, play "Des-pah-see-toe" (ph).
COMPUTER-GENERATED VOICE: Sorry. I don't know how to help with that.
SHAHANI: I try again with the song and the name of the artist. I try to spell it out - literally spell.
Play "D-E-S-P-A-C-I-T-O" (ph).
She totally doesn't understand. And then maybe because I'm asking so much, she gets desperate, throws another artist at me.
COMPUTER-GENERATED VOICE: OK. Here's Justin Bieber on Spotify.
(SOUNDBITE OF SONG, "FRIENDS")
JUSTIN BIEBER: (Singing) I was wondering about your mama.
SHAHANI: (Laughter) No, Justin Bieber is no Luis Fonsi. What's wild about this is I'm not asking for some obscure song Google's never heard of. "Despacito" is the single most-viewed video this year on YouTube, which is owned by Google.
PARAG CHORDIA: It probably was just out of vocabulary. It wasn't in the computer's lexicon, in the vocabulary of words that it knew about.
SHAHANI: Parag Chordia, an expert in artificial intelligence and speech, presents a theory.
CHORDIA: And so when it was trying to figure out what that sound was, it didn't know what to map it to.
SHAHANI: Google declined to discuss this failure of the home assistant. So I asked Chordia to make an educated guess. He says that even when a song is a YouTube hit, it may not automatically become part of the Google Home dictionary. Words need to be added by humans or by software. He's often had the experience of asking for some Indian classical singer Google Home doesn't know. For example...
CHORDIA: Tidaquwa - I have said Tidaquwa and that it's - and it's - and you know, it's giving me Shakira.
SHAHANI: Now, it's a great theory. Problem is, I debunked it. I established that my Google Home knows the word.
OK, Google, do you understand Spanish?
COMPUTER-GENERATED VOICE: Yes, I speak Spanish.
SHAHANI: OK, Google, why don't you understand what I'm saying in Spanish?
COMPUTER-GENERATED VOICE: I'm still learning it, but I can translate for you. Just ask.
SHAHANI: OK, Google, translate despacito.
COMPUTER-GENERATED VOICE: Slowly.
SHAHANI: Yes. She wasn't telling me to say it slowly. That's literally what despacito means. And after asking 23 times, here's what finally happened.
OK, Google, play "Despacito."
COMPUTER-GENERATED VOICE: Sure - "Despacito" by Luis Fonsi.
(SOUNDBITE OF SONG, "DESPACITO")
LUIS FONSI: (Singing in Spanish).
SHAHANI: If it weren't in her Spanish vocabulary, this small miracle would not have been possible. So here's theory two by Chordia. Google Home needs training. Even if YouTubers want the song, maybe the people using this new device aren't asking for it as much. So more people have to ask and have to ask in different accents to make the computer's ear more responsive.
CHORDIA: So that it knows that, hey, people don't just say this word this way; they also say it this way. And so it just needs examples.
SHAHANI: Now, maybe this sounds unlikely given what a megahit we're talking about. For what it's worth, Chordia himself does not believe Google has some agenda against Latin music.
CHORDIA: (Laughter) I don't think so given that it's been played I think four and half billion times on YouTube.
(SOUNDBITE OF SONG, "DESPACITO")
FONSI: (Singing in Spanish).
SHAHANI: Aarti Shahani, NPR News, San Francisco.
(SOUNDBITE OF SONG, "DESPACITO")
FONSI: (Singing in Spanish).
SHAHANI: OK, Google, thank you.
COMPUTER-GENERATED VOICE: You're welcome.
Copyright © 2017 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.