Lots of work is going into voice-to-text applications for hand-held devices. Translation from one language to another using a cloud-based system is one application I’ve seen demonstrated recently.
Such devices could help those who cannot hear make some sense out of a conversation with someone whose sloppy lips are hard to read and they’ll definitely help with phone conversations, but those of us who ever misunderstood the tone of an email or text message and ended up in a flame-war can attest to the imperfection of hand-crafted written language. These machines won’t help detect sarcasm.
Modern, low bandwidth codecs, like those used for Voice over IP, don’t actually send the original sound. They break the part of their input that they detect to be human voice into smaller parts and then approximate the tonal qualities on the other end. It’s purely digital mumbo-jumbo — if you look at the waveform coming out, it’s not really much like what went in, but your brain thinks it hears your buddy.
Your telephone uses an approximation called pulse code modulation (PCM) that takes a 128kbp digitized sound and then just ignores frequencies that aren’t often part of human speech, only sending 56kbps. But, it’s analog… basically what goes in, comes out. Even music sound almost fine over the phone, except a lot of the frequency range is missing.
I’m sure you’ve noticed that the codec in your cellphone does a horrible job of sending music. It’s even worse for tighter codecs on VoIP sending like 8kbps.
But the codecs are good enough for being able to tell if a friend is kidding or serious, being sarcastic, or dry, or both.
So, what if you hooked up one of these codecs with an arm band that vibrated a little transducer thing along with the voices that it picks out of the room. Seems like it would be less confusing that just jiggling a person’s arm along with all the noise in the room.
Wouldn’t the human brain eventually learn to discern verbal information from the vibrations?
No, I’m not going to work on it.
Leave a Reply