Listen Up! How Recognizing Sounds Can Help With Car Repairs
AILSA CHANG, HOST:
There are some people who are very good at recognizing sounds and replicating them, in the case of NPR's Car Talk guys.
UNIDENTIFIED MAN: Sticky lifters - (imitating car noises). Boiling coolant - (imitating car noises).
CHANG: So a tech startup is trying to do it for us. It isolates the sounds our cars make when they need an oil change or have sticky lifters, just like having the Car Talk guys in your own garage. Liz Gannes reports.
LIZ GANNES, BYLINE: Cars that make weird sounds are very normal. Sebastien Christian is a Silicon Valley scientist who researches hearing, and he was talking to an automaker about this.
SEBASTIEN CHRISTIAN: They realize that 50 percent of customers arriving in a garage, a repair shop, had the same problem, which was, I have a sound in my car. I have a sound a car in my car I can't identify, and I don't know what it is.
GANNES: We humans are bad at getting our machines to tell us what's wrong with them even though they kind of can, if only we could understand.
CHRISTIAN: You have a machine at home, a drill. And you know that, well, usually, after a year, the drill is about to break, so you change it.
GANNES: Or like my car, I take it in every 6,000 miles.
CHRISTIAN: Yes, exactly. And maybe your car could have waited until 8,000 miles, or you should have brought it at 5,000 miles. And the idea is to adjust that dynamically.
GANNES: By listening. If we could know what's going on with the car by the sound it makes, we'd be able to tell exactly when it needs service. Of course, sometimes you can also see when something breaks, but our ears would help our eyes with car repair. Sebastien Christian is a quantum physicist and neuroscientist who spent much of his career studying deaf children. He modeled how human hearing works. And then he realized, hey, I could use this model to help other deaf things, like, say, almost all machines.
Now, Christian has a startup called OtoSense doing just that in Palo Alto, Calif. He's got to deal with Peugeot in Europe to put microphones in auto shops, microphones that know the meaning of up to 65 different sounds. And sound recognition could be helpful for more than just hearing the aches and pains of cars.
CHRISTIAN: Constantly, everywhere, sound is bringing relevant information.
GANNES: So OtoSense, which has raised millions of dollars, is working with companies like Samsung on sensing sounds inside homes and sensing sounds inside everyday tools. Take a handheld drill. Christian plays the sound on his laptop. If you know that when you run it at a high RPM...
CHRISTIAN: Screw full speed.
(SOUNDBITE OF DRILL)
GANNES: ...It overheats and dies.
(SOUNDBITE OF DRILL BREAKING)
CHRISTIAN: Doesn't sound good.
GANNES: You could program it to automatically shut down when a tiny 5 cent microphone hears that th-th-th-thunk (ph). Or it could be the city itself that hears, not just the cars. If OtoSense microphones were built into stoplights, you could...
CHRISTIAN: Turn the light green when you hear a fire truck siren arriving.
(SOUNDBITE OF SIREN)
GANNES: Christian understands the scary potential of what he's building. He told me if he listened to me typing...
(SOUNDBITE OF KEYBOARD TYPING)
CHRISTIAN: After 3 hours, we can reconstruct 50 percent of what you're typing just by listening.
GANNES: Because my way of typing each key has its own sound, so his software could start to figure out each number and letter and question mark and period. Christian is determined to build the stuff ethically, to design microphones that detect certain auditory cues, not listen to every word people say. For NPR News, I'm Liz Gannes.
(SOUNDBITE OF TROMBONE SHORTY'S "TRIPPED OUT SLIM")
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. 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.